Aalborg University The Faculty of Engineering and Science Electronics and Electrical Engineering - Department of Acoustics Fredrik Bajers Vej 7
d
DK-9220 Aalborg East
Title:
Loudness Perception of Impact Noise with Equal Energy - evaluation of B-duration, peak SPL and frequency content Acoustics February 1st - June 1st 2006
Theme: Project period:
Abstract Project group: ACO1060 Group members: Andreas Gregers Gregersen Lars Sommer Søndergaard
Supervisor: Rodrigo Ordonez Co-Supervisor: Miguel Angel Aranda de Toro
Publications: Pages:
9 82
Impulsive signals consisting of brief bursts of acoustic energy, distributed in a short period of time are inaccurately assessed by the standards regarding hearing impairment. The aim of this thesis is to find physically descriptive parameters for impact noise, and evaluate the influence of these in terms of perceived loudness. Measurements of realistic impulsive noise, created by combinations of impacts between wood and metal have been performed in controlled environments at the Department of Acoustics at Aalborg University. Based on these measurements it is chosen to investigate the parameters: Bduration, peak SPL and frequency content, in terms of loudness. To evaluate the loudness perception caused by impact noise with variance in the given parameters, a subjective 2IFC listening test is performed. The results from this test showed no obvious relation between perceived loudness and the difference in B-duration, even though some compared stimuli were perceived with a significant difference. The test of the influence of the peak level, at least for the levels used in this thesis, showed results similar to the B-duration results, meaning no obvious relationship. The results from the test of the frequency content showed a significant perceived difference in loudness between stimuli based on a metal plate compared to a wooden plate, where the metal stimuli were perceived as the loudest.
Preface This master thesis is written by group ACO1060 on the 10th semester at the Department of Acoustics at Aalborg University. The title of the thesis is: “Loudness Perception of Impact Noise with Equal Energy - Evaluation of B-duration, Peak SPL and Frequency Content”. The thesis is documentation for the learning process accomplished for this semester, together with the theories and methods used for solving the problem. The documentation consists of a main report with associated appendices where topics from the main report are elaborated. The appendices are collected in a separate appendix document with bibliography and table of contents. The work on the project was carried out during the period from 1st of February to 1st of June 2006. In the report, cite references are written in squared brackets, e.g. [Moore, 2004, p. 100]. If no reference to a page number or chapter is present inside the brackets, it means that the information is taken from several locations in the literature. Further information about the reference can be found in the bibliography for the main report on page 82 or in the bibliography for the appendices in the appendix report on page 81. On the inside of the back cover of the main report is attached a DVD which contains this report in PDF format, MATLAB code for running the listening test with associated files and un-normalized recorded signals. Contents of the DVD: - PDF file of the project - MATLAB code for listening test - Recorded signals
II
———————————–
———————————-
Andreas Gregers Gregersen
Lars Sommer Søndergaard
Contents 1
Introduction 1.1 Aim of This Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2
Impulsive Noise 2.1 Physical Properties . . . . . . . . . . . . 2.2 Psychophysical and Physiological Aspects 2.3 Assessment of Noise . . . . . . . . . . . 2.4 Summary . . . . . . . . . . . . . . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
3 3 9 14 20
Analysis of Measurements 3.1 Choosing Signals for Recording 3.2 Recording of Impact Noise . . . 3.3 Analysis of Recordings . . . . . 3.4 Analysis of Parameters . . . . . 3.5 Choice of Parameter . . . . . . 3.6 Listening Test Stimuli . . . . . .
3
4
5
6
1 2
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
21 21 22 24 27 38 40
Listening Test 4.1 Methods for Testing the Hypothesis 4.2 Design of Listening Test . . . . . . 4.3 Test Subjects . . . . . . . . . . . . 4.4 Reproduction of Test Files . . . . . 4.5 Procedure for Selecting Test Files. . 4.6 Execution of the Listening Test . . . 4.7 Analysis of Presented Stimuli . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
43 43 44 46 46 47 49 51
Evaluation of Results from Listening Test 5.1 Statistical Method . . . . . . . . . . . . . . 5.2 Statistics on B-duration . . . . . . . . . . . 5.3 Statistics on Peak Level . . . . . . . . . . . 5.4 Statistics on Frequency Content . . . . . . 5.5 Results Across Energy Normalizations . . . 5.6 Summarizing and Discussion of the Results
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
59 59 61 66 70 74 75
Conclusion 6.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
77 79
Bibliography
80
. . . . . .
III
C HAPTER Introduction
1
Exposure to noise has for several years been known to cause hearing impairment and thereby reduction of quality of life. Limits for noise exposure have therefore been constructed together with standardized assessment methods in order to prevent additional hearing impairment. This is especially a known problem in industry with different types of noise. People, who work in the metal industry, where impulsive noise is often experienced, however still experience hearing impairment even when the limits for noise exposure are not exceeded. This problem is described in the article [Arbejdsmiljø, 2002] together with a description of a Dutch investigation which discovered that the noise level could be raised 12 dB for workers in the wood industry before they experienced the same amount of hearing impairment as workers in the metal industry when exposed to impulse noise. These investigations indicate that the current methods for evaluating the danger of impulse noise are not sufficient. These methods are mostly based on calculations of the equivalent A-weighted continuous sound pressure level, LAeq , which describes the energy integrated over a certain period of time with a specific weighting. Impulsive noise are a complex hazardous type of noise which is hard to measure due to its brief high peak level bursts of acoustic energy distributed in a short period of time. The typical duration is much shorter than the time window used for calculating the LAeq which therefore are not able to properly describe the impulses but rather judge the exposure as lower than it really is. In the current legislation about noise assessment there is very little information about how to treat sounds with impulsive characteristics. In general if such sounds are identified a simple penalty of at least 5 dB is added to the LAeq level depending on the type of impulsive noise. There is very little information as to the relation between these suggested corrections and the human perception (i.e. loudness or annoyance) of noise of tonal or impulsive character. More knowledge about impulsive noise is important since people are subjected daily to impulsive noises mostly without noticing it. Since impulsive sound only rarely occurs in everyday life this exposure are not dangerous. Impulse noise at work is more dangerous. Many factories have machines which emit impulse noise, it could be stamping machines etc. At construction sites there is also a huge amount of noise based on impacts, e.g. hammering. A third dangerous place is in the military where explosions and gunfire also emit impulsive sound. At these places persons can be subject to these impulsive noises in a working day.
1
b
CHAPTER 1. INTRODUCTION
Many of the methods used in the assessment of noise utilize frequency weighting functions which are meant to judge a given sound in a way equal to how loud such a sound is perceived by the human auditory system. To try to deal with these problems, methods that are more descriptive of the physical properties of impulsive signals should be found, together with information about how this impulsive noise is perceived by the auditory system in terms of loudness.
1.1
Aim of This Thesis
The aim of this thesis is to evaluate different parameters descriptive of impulsive noise in terms of perceived loudness in relation to the equivalent continuous frequency weighted sound pressure level. The parameters shall be chosen based on an analysis of measured types of impulsive noise. The perceived loudness sensation evoked by changes in the found parameters shall be evaluated based on a subjective listening test.
2
C HAPTER Impulsive Noise
2
Impulsive sound is “sound characterized by a brief bursts of sound pressure” and “The duration of a single impulsive sound is usually less than 1 s.” [ISO 1996-1, 2003]. As seen from this description impulse sound is a broad definition, embracing wide areas of different sounds. Most characteristic for this type of signals are sounds like gunfire, explosions, fireworks and hammering on metal. The aim of this chapter will be to describe the aspect of impulsive sound, including a description of its physical properties, parameters which can be used to describe impulsive sound, physiological aspects related to impulsive sounds and assessment of impulsive noise.
2.1
Physical Properties
This section will consist of a description of the physical properties of impulsive noise, along with a description of where impulsive noise occurs and relevant parameters used in literature and standards to describe impulsive sounds.
2.1.1
Description of Impulse Signals
An ideal impulse is in discrete time defined as a one at zero time and zero for the rest of the time, this is also referred to as the Dirac delta function [Oppenheim & Schafer, 1999, p. 11]. Impulsive noise will more or less have similar properties with the ideal impulse dependent on how much they resemble each other. In the following sections different aspects which have an influence on the characteristics of impulsive sound will be described.
Generation of Impulsive Noise The generation of impulsive noise is generally divided into two categories: impulse noise and impact noise [Hamernik & Hsueh, 1991].
3
b
CHAPTER 2. IMPULSIVE NOISE
• Impulse noise is mainly created by explosives and is characterized by: “a noise transient that arises as the result of a sudden release of energy” and peak SPL can often be more than 194 dB. Examples are explosions, gunshots, fireworks, etc. • Impact noise is created by impacts, and is characterized by: “a noise transient that arises as the result of the impact between two objects”. An example would be when a hammer strikes a metal plate. A time plot of examples of each type is shown in figure 2.1 together with a time plot of the ideal impulse. Notice that the y-axis are different for each plot. Ideal impulse
Gun shot
2
Metal impact
600
1.5
Wood impact
50
50
40
40
30
30
20
20
400 1
−0.5
0
Sound Pressure [Pa]
0
Sound Pressure [Pa]
Sound Pressure [Pa]
Sound Pressure [Pa]
200 0.5
10
0
−10
10
0
−10
−200 −20
−20
−30
−30
−40
−40
−1 −400 −1.5
−2 −50
0
50 100 Time [ms]
150
200
−600 −50
0
50 100 Time [ms]
150
200
−50 −50
0
50 100 Time [ms]
150
200
−50 −50
0
50 100 Time [ms]
150
200
Figure 2.1: Time plots of: 1: Ideal impulse, 2: Gun shot recorded in the anechoic room in a distance of approximately 3 m, 3: Impact of wood weight on metal plate recorded in the anechoic room in a distance of 2.61 m, 4: Impact of wood weight on wood plate recorded in the anechoic room in a distance of 2.61 m. The used gun is a Röhm RG76 6mm with FLOBERT PLATZ ammunition, AAU number 229. For more information about the recordings of the impact sounds see appendix B.
When comparing the chosen examples of impulse sounds and impacts sounds in figure 2.1 it is seen that they are very different. The un-weighted peak SPL for the gun shot is 149 dB, where the un-weighted peak SPL for the impact sounds are 128 dB and 121 dB. See definition of peak SPL in equation (2.1). The waveform is also very different where the waveform of the impulse sound is similar to the ideal impulse, while the impact sounds has a much longer decay time, which is due to vibrations in the material. In figure 2.2 is shown the same signals as displayed in figure 2.1 but with a shorter time window to have a closer look at the waveform at the time of the peak value. It is then possible to observe that the waveform of the gun shot practically only has a duration of 1 ms which consist of a very fast rise to the (negative) peak value followed by a positive peak value, after only one transition steeply decays to the initial position (≈ 0 Pa). At 2 ms can be seen a reflection, due to grid and platforms in the anechoic room. For the impact sounds it can be seen that the path from the initial positon to the peak value followed by the decay to the initial position is not as direct as with the impulse sound but with numerous oscillations.
Typical Peak Values for Impulsive Noise The impulsive noises normally treated in literature are those with the highest peak SPL, since these levels are presumed to create most hearing damage. However bursts with low peak SPL are also impulse noise. 4
2.1. PHYSICAL PROPERTIES
Ideal impulse
Gun shot
2
Metal impact
600
1.5
Wood impact
50
50
40
40
30
30
20
20
400 1
−0.5
0
Sound Pressure [Pa]
0
Sound Pressure [Pa]
Sound Pressure [Pa]
Sound Pressure [Pa]
200 0.5
10
0
−10
10
0
−10
−200 −20
−20
−30
−30
−40
−40
−1 −400 −1.5
−2 −2
0
2 Time [ms]
4
6
−600 −2
0
2 Time [ms]
4
6
−50 −2
0
2 Time [ms]
4
6
−50 −2
0
2 Time [ms]
4
6
Figure 2.2: Same plots as figure 2.1 but with a shorter time window.
Normally impact noise is considered to have peak SPL values below 140 dB, and this value has normally been considered to be the fence between impact noise and impulse noise. It has also been shown that nonlinear wave behavior starts to affect the appearance of the waveform of the noise at approximately 140 dB for mid frequency pure tones. Below 140 dB there is not much difference in the wave profile between impact noise and impulse noise, and no distinction should be made between them [Hamernik & Hsueh, 1991].
Spectral Contents For an ideal impulse the frequency contents will be equally distributed from -π to π [Oppenheim & Schafer, 1999, p. 62]. From this can be concluded that the more similar the waveform is to the waveform of the ideal impulse the more equally will the frequency contents be distributed. This is illustrated in figure 2.3 which shows time frequency plots of each of the plots displayed in figure 2.1. Each plot is normalized to the peak value, which is set to 0 dB to ease the comparison. The frequency content for the ideal impulse is equally divided as expected. The gun shot is approximately ideal with a broad frequency range. The impact plots have also a broad frequency content for the time of the impact.
Environment The environment also has a high influence on the final waveform, especially for the decay of the sound pressure. The more reverberation the longer the decay, since each reflection add an attenuated delayed copy of the original sound. The acoustic properties of the environment also have a high influence on the waveform in terms of frequency response, where room modes will be present and specific frequencies will therefore be attenuated or amplified.
2.1.2
Occurrence of Impulse Sounds
Impulse sounds occur frequently in everyday life, and are sounds like slamming of doors, sounds of church bells ringing, clapping with hands and cheerful applause at sport games can also be denoted as impulsive sounds. Those mentioned here are however mainly occurring once and a 5
b
CHAPTER 2. IMPULSIVE NOISE
Gun shot
Metal impact
Wood Impact 20
18
18
18
18
16
16
16
16
14
14
14
14
12
12
12
12
10
8
10
8
Frequency [kHz]
20
Frequency [kHz]
20
Frequency [kHz]
Frequency [kHz]
Ideal impulse 20
10
8
10
8
6
6
6
6
4
4
4
4
2
2
2
2
−50
0
50 100 Time [ms]
150
200
−50
0
50 100 Time [ms]
150
200
−50
0
50 100 Time [ms]
150
200
−50
0
50 100 Time [ms]
150
200
Colorbar
−70
−60
−50
−40
−30
−20
−10
0
Figure 2.3: Spectrograms of the plots displayed in figure 2.1. Each plot is normalized so their un-weighted peak SPL is 0 dB (black). Values from 0 dB to -70 dB is displayed.
normal part of everybody life so it is not considered as noise. A more dangerous and annoying occurrence of impulse sounds happens in the industry and in the military since both total energy and peak level are higher and the impulsive sounds occur often. In the industry numerous companies have machines which emits impact sounds, for example in printing companies, where the printing press emits periodic impact noise. Constructor sites are also a place where workers are highly exposed to impact noises, for example hammering and nail guns. For the military the impulse sounds will be due to an explosion of some kind; firing of guns, firing of missiles, explosions of explosive, etc. The impact noise and impulse noise generated in industry and in the military will not only be single sound events but also repeated impulsive signals are likely to occur. This is highly depending on whether the generator of the noise is automatic the noise will be periodic, and if it is manual the duration between each noise burst will be randomly, but might still have some kind of periodicity. Most impact noise has it origin in the industry, and will mostly be repeated in series of impulses. This impulse series will have some degree of periodicity, since most industrial processes are automated. The part of impulse noise that originates from the military will normally also be repeated, and the periodicity will again depend on if the weapons or sound sources are manual or automatic. In this thesis it is chosen to concentrate on only one repetition of impulsive noise; meaning that for periodic impulsive noise only one occurrence will be investigated at the time.
2.1.3
Parameters describing Impulsive Noise
In this section is listed some of the parameters or metrics often used in literature to describe noise, and especially impulsive noise.
6
2.1. PHYSICAL PROPERTIES
Peak Sound Pressure Level The most common descriptor of impulsive noise is peak sound pressure level, C-weighted or unweighted. Peak SPL is defined as: “ten times the logarithm to the base 10 of the ratio of the square of the peak sound pressure to the square of the reference sound pressure, where the peak sound pressure is the maximum absolute value of the instantaneous sound pressure during a stated time interval with a standard frequency weighting or measurement bandwidth” [ISO 1996-1, 2003], see equation (2.1):
Lpeak = 10 log10
pp po
2
[dB]
(2.1)
where p p is the peak sound pressure in Pascal and po is the reference sound pressure, which in air is 20 µPa. According to [ISO 1996-1, 2003] C-weighting should normally be applied, however in the area of impulsive noise studies several litterature as [Hamernik & Hsueh, 1991] and [Coles et al., 1967] uses unweighted peak SPL which will also be used in this thesis. The different weightings will be specified in section 2.3.2.
Equivalent Continuous Sound Pressure Level Another common descriptor of noise is the equivalent continuous sound pressure level, A-weighted or un-weighted. Equivalent continuous sound pressure level is defined as: “ten times the logarithm to the base 10 of the ratio of the square of the root-mean-square sound pressure over a stated time interval to the square of the reference sound pressure, the sound pressure being obtained with a standard frequency weighting” [ISO 1996-1, 2003], see equation (2.2):
LAeqT = 10 log10
1 T
Z
T
p2A (t)/p2odt
[dB]
(2.2)
where pA (t) is the A-weighted instantaneous pressure at time t, p0 is the reference sound pressure (20 µPa) and T is the integrating time, often Fast (F = 125 ms) or Slow (S = 1000 ms).
Sound Exposure Level Sound exposure level is defined as: “ten times the logarithm to the base 10 of the ratio of the sound exposure, E, to the reference sound exposure, E0 , the sound exposure being the time integral of the time-varying square of the frequency-weighted instantaneous sound pressure over a stated time interval, T , or an event” [ISO 1996-1, 2003], see equation (2.3):
LE = 10 log10
E E0
dB
(2.3)
7
b
CHAPTER 2. IMPULSIVE NOISE
where E=
Z
p2 (t)dt
dB
(2.4)
T
and E0 is equal to the square of the reference sound pressure of 20 µPa multiplied by the time interval of 1 s.
Rise Time The rise time (Tr ) of a impulse describes how fast a impulse rises. At the figure 2.4(a) the rise time (Tr ) is shown, as the time difference between the initial resting position and the peak. This is valid since this is an idealized plot. According to [DS/ISO 10843, 1998] signal rise time is defined as: “Time, in seconds, a signal takes to rise from 10 % to 90 % of its maximum absolute value of the sound pressure.” T
B
p(t)
p(t)
20 dB
t
t
Tr −−−−TA−−−−
(a)
(b)
Figure 2.4: Rise time, Tr , and fall times. (a): A-duration, TA and (b): B-duration, TB .
A-duration and B-duration In [DS/ISO 10843, 1998] two durations are used to describe impulsive noise, named A- and Bduration. A-duration is mostly used for describing impulse noise, where B-duration is mostly used for describing impact noise [Henderson & Hamernik, 1986]. In figure 2.4(a), the A-duration, TA is shown. It is defined as the time it takes to go from the initial resting position (sound pressure = 0) and to cross zero again after the impulse peak. In practice instead of zero a value 20 dB down from the peak is used, which is also equal to 10 % of the peak sound pressure. In figure 2.4(b), the B-duration, TB is shown. For calculating the B-duration, the envelope of the un-weighted impulse is found. TB is then the time difference between the two points where the envelope crosses a value 20 dB lower than the peak value, referred to as the -20 db limit. If more impulses are present in the signal the B-duration is a summary of the B-duration for each single impulse. The definition of the envelope is weakly defined in the standard. In appendix A is described how B-duration will be calculated for this project.
8
2.2. PSYCHOPHYSICAL AND PHYSIOLOGICAL ASPECTS
Kurtosis Kurtosis is a statistical measure used to evaluate distributions and gives the relation between the amount of samples close to the peak value and the amount of samples close to the mean value. Kurtosis is defined as the ratio between the fourth moment (m4 ) and the squared of the second moment (m2 ), [Erdreich, 1986].
β=
m4 ∑(xi − x)4 = n (m2 )2 (∑(xi − x)2 )2
(2.5)
where moment is defined as: mr =
1 (xi − x)r n∑
(2.6)
where mr is the moment, n is the number of samples and r is the order of the moment. For a signal represented as pressure over time, like an acoustic signal, the variable xi can be represented as the instantaneous pressure p(t) and for signals without any DC-component the mean value can be assumed to be zero. This gives: m4 = β(t) = (m2 )2
1 RT T 0 1 RT T 0
p4 (t)dt p2 (t)dt
(2.7)
Crest Factor Crest Factor (CF) is defined as the ratio between the peak pressure (Pmax ) and the root-mean-square pressure (Prms ), [Erdreich, 1986]: CF =
Pmax Pmax =√ Prms m2
(2.8)
Physical properties of impulsive sound have now been investigated together with relevant parameters and perceived effects of impulsive sound should be analyzed.
2.2
Psychophysical and Physiological Aspects
The exact effects of impulsive noise on humans are not fully understood; how this noise is perceived and interpretated is therefore still a issue of research. This section will try to summarize on the theory related to psychophysical and physiological aspects of noise.
2.2.1
Perception of Loudness
Loudness of a sound is a subjective quantity and therefore cannot be measured through physical objective measurements but has to be evaluated through subjective listening tests in order to determine the perceived loudness of a given sound. According to [Moore, 2004] “loudness is defined as 9
b
CHAPTER 2. IMPULSIVE NOISE
that attribute of auditory sensation in terms of which sounds can be ordered on a scale extending from quiet to loud.” A topic that relates to the loudness perception of sounds is the equal-loudness-level contours which describe how loud sounds with different frequencies are perceived compared to each other for people with normal hearing.
Equal-Loudness-Level Contours The equal-loudness-level contours are based on the loudness level of pure tones. Each contour express the relation in loudness level between pure tones with different frequency and the corresponding sound pressure level (SPL). The equal-loudness-level contours are based on the concept of loudness level. To obtain the loudness level of a test sound consisting of a pure tone this sound must be compared in a subjective listening test to a reference sound. This reference sound is defined as a sound with a frequency of 1000 Hz propagating in a free-field environment perceived binaurally from the frontal direction. During the test to find the loudness level of the test sound the level of the reference sound is adjusted until the test sound and the reference sound are perceived as equally loud. The sound pressure level of the 1000 Hz reference sound is then the loudness level of the test sound expressed in phons. A test sound that is perceived to be as loud as the reference sound at 60 dB SPL is thus defined to have a loudness level of 60 phon. By definition a pure tone of 1000 Hz is perceived with a loudness level equal to the sound pressure level. Alternatively the 1000 Hz reference sound can be fixed in level and the test sound adjusted until they are perceived equally loud. This method is known as a loudness matching. When different frequencies covering the audible frequency range are used as test sounds a curve can be generated that describes which sound pressure level tones with different frequencies should have in order to be perceived with the same loudness level. These curves gives the connection between loudness level, sound pressure level and frequency, and are called equal-loudness-level contours or just loudness contours. The equal-loudness-level contours as defined in [DS/ISO 226, 2004] (binaural free-field listening and frontal incidence) can be seen on figure 2.5. The dashed curve is the Minimum Audible Field(MAF) curve which describes the threshold of hearing for binaural listening. Besides describing the loudness level of sounds with different frequencies, the equal-loudnesslevel contours are also the foundation for the frequency weighting functions, see section 2.3.2.
Loudness in Sones Sometimes it can be practical to be able to evaluate perception of loudness in terms of a single value. The units of loudness is sone and one sone is equal to 40 phon. A doubling of the perceived loudness of a sound with 1 sone is 2 sone and roughly corresponds to a 10 dB increase in the sound pressure level, which means that a loudness of 2 sone is equal to a loudness level of 50 phon. This relationship is only valid for levels above 40 phon.
10
2.2. PSYCHOPHYSICAL AND PHYSIOLOGICAL ASPECTS
Normal equal−loudness−level contours
120
100 phon
100
90 phon 80 phon
80 SPL [dB]
70 phon 60 phon
60
50 phon 40 phon
40
30 phon 20 phon
20
10 phon MAF 0 16
31.5
63
125
250
500 1000 Frequency [Hz]
2000
4000
8000
16000
Figure 2.5: Equal-loudness-level contours as defined in [DS/ISO 226, 2004]. The dashed line is an approximation to the hearing threshold (MAF). The 10 phon and 100 phon curves are dash dotted due to the lack of experimental data used for the definition for these curves.
The conversion between phon and sone for loudness levels above 40 phon can be performed using the equation (2.9). The relation between loudness and loudness level is according to [DS/ISO 532, 1975] given as: S=2
P−40/10
(2.9)
where S is loudness [sone] and P is loudness level [phon]. For further information about conversion from 1/3 octave band SPL to sones see [DS/ISO 532, 1975]. For loudness levels below 40 phon the loudness changes more rapidly as a function of loudness level. The loudness perception of sound is not only valid for pure tones as used for the definition of the equal-loudness level contours. This is due to the fact that the auditory system is capable of perceiving much more complex sounds with a broader spectrum and thereby also perceives a loudness sensation evoked by these sounds. Some of the parameters known to have a influence on the loudness perception are the bandwidth and the duration of the signal.
The Influence of Bandwidth and Duration on Loudness It is a common assumption, according to [Moore, 2004], that loudness is related to the total neural activity caused by a sound. This means that the loudness of a sound may depend on a summation of the total neural activity across different frequency channels, known as the critical bands. Given a complex sound with a bandwidth of W and with fixed energy then two situations can be imagined: Firstly, if the bandwidth W is less than a certain bandwidth called the critical bandwidth for loudness (CBL ), then the sound will be perceived as loud as a pure tone or narrowband noise 11
b
CHAPTER 2. IMPULSIVE NOISE
with equal intensity and frequency close to the center frequency of the given critical band, independent of W . Secondly, if the bandwidth W of the complex sound is increased beyond the CBL the perceived loudness of this sound will begin to increase. This is valid both for bands of noise and also complex sounds consisting of simultaneously pure tones. Since the process of listening is not discrete but rather a continuous ongoing process, the influence of the duration of the sound has an effect on the perceived loudness. According to [Moore, 2004] there is a general agreement that for a sound with a given intensity the perceived loudness will increase with a increased duration of this sound up till a duration of 100-200 ms. For durations up till 80 ms constant energy leads to constant loudness. It can sometimes be of interest to be able to calculate the loudness sensation evoked by a complex sound. This calculation can be performed using a loudness model. The most commonly used method is Zwickers loudness model which has been adapted in the [DS/ISO 532, 1975]. This method is based on the spectrum of the noise calculated in one-third octave bands. A drawback of this model is that it is only designed for stationary sounds. A newer loudness model has been proposed by [Moore et al., 1997]. This loudness model has been further improved by [Glasberg and Moore, 2002] to also be able to predict the loudness of time varying sounds.
Loudness of Impulsive Signals Loudness is, through the equal loudness level contours, well defined in the frequency domain for steady sounds, but for sounds with shorter duration or impulsive nature, the perceived loudness is not as precisely defined. According to [Moore, 2004] the loudness of sounds depends on their duration, and for short duration sounds their loudness may roughly depend on their total energy. Studies of subjective and objective measurement of the loudness level of single and repeated impulses from 1970 have been performed by [Reichardt, 1970]. He concludes that for tests concerning the loudness level, or loudness compared to a reference sound, it is of great importance that the compared signals are of equal duration. Otherwise the uncertainty of judging the loudness level of two compared sounds increases with an increasing difference in duration between these sounds. Due to this difficult task the subject might involuntary compare the duration of the sounds rather than the loudness level causing the signal with the longest duration to get a higher loudness assigned. In 1972 N. L. Carter studied the effects of rise time and repetition rate on the loudness of acoustic transients, [Carter, 1972]. As stimuli in this study, artificially created triangular transients were used as a simplification of a impulse signals. All stimuli had a duration of 1 ms and varied rise times of 0.05, 0.10, 0.25 or 0.50 ms and repetition rates from 1 to 256 pulses per second. From these studies Carter concluded that “... the rise time of the individual pulses and repetition rate significantly affect the loudness of pulse trains”. The effect of the rise time was that an increased rise time requires the level of the pulses to be increased in order to maintain the same loudness level. The effect of the repetition rate confirmed the results of a earlier study from 1963 by the same author, [Carter, 1965], which indicated “... that for each doubling of the repetition rate a reduction of 3 dB in peak level of the transients is required to maintain equal loudness”.
12
2.2. PSYCHOPHYSICAL AND PHYSIOLOGICAL ASPECTS
2.2.2
Noise Induced Hearing Loss
Noise, and especially noise with high SPL, is known to cause damage on the human auditory system and thereby causing hearing loss. Within the last couple of decades the political focus on noise pollution in the community has been growing, both concerning noise at work but also in leisure time. The topic of noise exposure and how to assess this noise both in terms of measurements and the following abatement has grown even larger. One special type of noise that has gained some special attention is the impulsive noise. Some of the issues related to impulsive noise are: • Impulsive noise may be more hazardous than continuous noise since it may create direct mechanical damage to the cochlea just after a single exposure. In addition the recovery of the hearing after exposure to impulsive noise may be erratic and non monotonic. [Henderson & Hamernik, 1986]. • The equal energy hypothesis, which is commonly accepted for judging continuous noise, does not hold for impulsive noises, which has been shown by [Henderson et al., 1990]. • Impulsive noise present in continuous background noise can cause even greater hearing loss than what would be caused by the sum of the two individual noises presented individually [Henderson & Hamernik, 1986]. The problem with impulsive noise is a highly relevant topic since it is a known problem in the military regarding weapons discharge and explosions and maybe an even more relevant and common problem in the industry where hammering and stamping machines etc. create a hazardous working environment, which can cause Noise Induced Hearing Loss (NIHL). The direct connection between exposure to impulsive noise and NIHL in terms of the amount of lost hair cells in the cochlea is still a topic of research. In the struggle to find this connection D. Henderson et al. studied in 1990 the importance of level, duration and repetition rate in impact noise in relation to the equal energy hypothesis concerning hearing loss [Henderson et al., 1990]. To find this connection he studied the hearing threshold of several chinchillas measured through an electrode implant in the brain. The animal were exposed to broadband impact noise of 200 ms duration and with various peak sound pressure levels from 107 to 137 dB SPL and three different repetition rates but all with equal acoustic energy (LE ). The experiment showed that the Temporary Threshold Shift (TTS)1 in the chinchillas caused by these noise exposures were highly dependent on the peak level and the duration of the impulses. Furthermore the studies showed a clear connection between the Permanent Threshold Shifts (PTS)1 measured on the chinchillas and the amount of lost hair cells measured via light microscopy after the animals were executed. Finally the study showed that the equal energy hypothesis does not hold for exposure to impulsive noise. Since it is not ethically and morally accepted to do this kind of on-purpose-damaging experiments on humans, the effects of exposure to impulse noise have to be evaluated in a more human justi1 The minimum detectable level of a sound in the absence of other external sounds is known as the absolute threshold of a sound. After exposure to a noise the absolute threshold might have changed and this change (in dB) is called threshold shift. If this shift is temporary this is known as Temporary Threshold Shift (TTS) otherwise it is called Permanent Threshold Shift (PTS).
13
b
CHAPTER 2. IMPULSIVE NOISE
fiable manner. Tests with impulse noise (or tests with noise exposures in general) on humans are therefore often evaluated in terms of mild levels of TTS (20-25 dB) caused by exposure to noise. Due to these ethical concerns research concerning exposure of impulsive noise and the caused NIHL on humans cannot be carried out. Therefore it is still uncertain completely how to evaluate this “special” type of noise.
2.2.3
Effects of Noise in the Community
Besides the obvious danger in terms of NIHL caused by impulsive noise this type of noise (or noise in general) still has other documented physiological and psychological effects on the health of human beings. These other effects are: stress, sleep disturbance or even cardiovascular diseases. A review and discussion of these effects can be found in [Kryter, 1985]. One of the major noise sources contributing to the noise pollution in the community is traffic noise (road, train and aircraft noise). This noise causes annoyance to the inhabitants in every larger town and city since traffic noise is present close to homes and resting locations. The industry is another important source of noise, but this type of noise is normally most concerning the workers in the industry since larger factories are normally located in industrial quarters without residential property. To deal with this environmental noise issue the European Union has conducted a noise directive [Directive 2002/49/EC, 2002] that should be implemented in the nations legislation across the European Union. The directive is mainly concerned with traffic noise and industrial noise. The directive demands that all EU member nations makes noise maps, starting with major cities, roads, railroads and airports, and to make action plans for the abatement of this kind of noise in the relevant areas. Noise has been regarded as an un-ignorable health threat to the community. This has forced the World Health Organization (WHO) to attempt to deal with this problem by conducting some guidelines in the assessment of noise [Berglund et al., 1999]. The problems associated with noise are: noise induced hearing impairment, interference with communication, sleep disturbances, mental health effects, cardiovascular and psychophysiological effects, effects on performance, annoyance response and effects on social behavior. [Berglund et al., 1999] describes only guidelines for the assessment of noise and the purpose of the document is more to summarize the health issues related to noise exposure in the community and as a proposal for a debate rather than proposals for actual legislations.
2.3
Assessment of Noise
The way to asses and measure environmental noise is specified in [ISO 1996-1, 2003]. Which parameters of the measured noise should be calculated depends on the type of noise and the purpose of the noise measurement. One of the most common types of noise is the continuous noise. The sound pressure level of this type of noise can be constant, fluctuating or slowly varying over time. According to the [ISO 1996-1, 2003] these variations of noise is assessed by calculating the A-weighted equivalent
14
2.3. ASSESSMENT OF NOISE
continuous sound pressure level, see equation (2.2) in section 2.1.3. For noise sources with a fluctuating or varying level the maximum A-weighted sound pressure level with a time weighting, see equation (2.11), is also permitted to be used. The frequency weightings together with the time weightings are defined in section 2.3.2. The A-weighted sound pressure level with a time weighting (LAτ (t)) is defined as: ! 1/2 Z t 1 −(t−ξ)/τ 2 LAτ (t) = 20 log10 ( ) p (ξ)e dτ /p0 (2.10) τ −∞ A where τ is the exponential time constant expressed in seconds and is F for fast or S for slow, see section 2.3.2. ξ is a dummy variable used for the time intergration from −∞ to the time of observation t. p0 is the reference pressure (20 µPa) and pA (ξ) is the A-weighted instantaneous sound pressure. The maximum A-weighted sound pressure level is defined as seen in equation (2.11) LAτmax = max(LAτ ) [dB]
(2.11)
The impulse noise, characterized by its brief bursts of sound pressure is due to its complex nature hard to categorize and therefore more difficult to assess compared to continuous noise.
2.3.1
Assessment of Impulsive Noise
In the [ISO 1996-1, 2003] impulsive sound is divided into the following three categories: High-energy impulsive sound sources includes explosive sources with a equivalent mass of minimum 50 g TNT, or sound sources with comparable characteristics. Examples of this type of noise are: Quarry and mining explosions, sonic booms, demolitions, industrial processes that uses highly explosives and military ordnance. Highly impulsive sound sources are sound sources with highly impulsive characteristics and a high degree of intrusiveness. Examples of this type of noise are: Small fire arms, nail guns, hammering on metal or wood, pneumatic hammering, pavement breaking, punch presses or pile drivers. Regular impulsive sound sources are impulsive sound sources that are neither highly impulsive nor high-energy impulsive sound sources. Examples of this type of noise are: Slamming of a car door, church bells or applauses at outdoor ball games (e.g. soccer). The [ISO 1996-1, 2003] describes how to assess sound sources with a single event or repeated events; this includes single impulse noises as those concerning this project. According to the standard all single event sounds, except high-energy impulsive sounds and sounds with strong low frequency content, should be A-weighted unless another frequency weighting is recommended for the given measured quantity. The three preferred descriptors of single event sound sources are: • Sound exposure level, frequency weighted (LAE ), see (2.3) 15
b
CHAPTER 2. IMPULSIVE NOISE
• Maximum sound pressure level, frequency and time weighted (LAF max ), see (2.11) • Peak sound pressure level, frequency weighted (LCpeak ), see (2.1) The duration of an event should according to [ISO 1996-1, 2003] be specified relative to a known characteristic of the sound. For instance the number of times a certain fixed level is exceeded. Repetitive single events should be described by the A-weighted sound exposure level (except for high-energy impulsive sounds) of one single event and the number of times the event occurs or is repeated. Based on this and the type of sound event the adjusted sound exposure level (LREi j ) can be calculated, see equation (2.12).
LREi j = LEi j + K j
(2.12)
where LEi j is the measured sound exposure level of the ith single sound event and and K j is the level adjustment dependent on the jth type of noise, [ISO 1996-1, 2003]. Based on the adjusted sound exposure level and the number of times the single event was measured the rating equivalent continuous sound pressure level (LReq j,T n ) should be calculated. See equation (2.13).
LReq j,T n = 10 log10
1 L 10 REi j/10 ∑ Tn i
!
[dB]
(2.13)
where Tn denotes the time interval in which the repeated events occur, LREi j is the adjusted sound exposure level for the ith sound event for the jth type of noise. The value of K The correcting value for impulsive noise and noise with strong tonal character, according to [ISO 1996-1, 2003], can be seen in table 2.1. Source character
Adjustment (K) [dB]
Regular impulsive
5
Highly impulsive
12
High-energy impulsive
∗
Prominent tones
3 to 6
Table 2.1: Table showing the value of the correcting factor K dependent on the type of noise. ∗ depends on the Cweighted sound exposure level, see annex B in [ISO 1996-1, 2003] for details.
2.3.2
Time and Frequency Weightings
The time and frequency weightings have been adapted into the standards to make the assessment of noise similar to the way it is perceived in the human auditory system.
16
2.3. ASSESSMENT OF NOISE
The frequency weightings origins from the equal loudness level contours, see section 2.2.1, which describes the relation between how loud sounds with different frequencies are perceived compared to each other (to a 1 kHz tone). All the frequency weighting functions are designed to have a gain of 0 dB at the frequency of 1 kHz and are defined in the [DS/IEC 61672, 2003]. The frequency weightings function is widely used to objectively compare signals to the loudness perceived by the human auditory system. The A-weighting curve is based on the 40 phon equalloudness-level curve, see section 2.2.1, and was initially intended to be used with low level signals of approximately 40 phon. Today the A-weighting curve is widely used in the assessment of nearly all kind of noise such as environmental and industrial noise. The C-weighting curve is based on the 90 phon curve and is intended for higher level sounds, approximately 90 phon, but is today mostly used when measuring absolute peak level. The frequency characteristics for the A and C-weighting can be seen in figure 2.6. A and C−weighting characteristics 10
0
−10
Gain [dB]
−20
−30
−40
−50
−60
−70 A−weighting C−weighting −80 10
100
1000
10000
Frequency [Hz]
Figure 2.6: The frequency characteristics of the A and C-weighting functions as specified in the [DS/IEC 61672, 2003].
The time weightings are used in the standards to specify a standard integration time for sound level meters when measuring the time integration of a sound event. The two common weightings used for time integrating are specified in [DS/IEC 61672, 2003] as F and S. F is abbreviation for fast and is equal to a time constant of 0,125 s and S is abbreviation for slow and is equal to a time constant of 1 s.
17
b 2.3.3
CHAPTER 2. IMPULSIVE NOISE
Measuring Equipment for Impulsive Noise
The standard [DS/ISO 10843, 1998] describes the standardized way to measure physical parameters of single or repeated impulsive noises and the requirements for the equipment recommended for such measurements. Since impulsive noise consists of a certain amount of acoustical energy concentrated in a very small period of time this results in some tight demands for the measuring system, both concerning the time dependent specifications and those dependent on the level of the signal. The requirements for the measuring systems depend on the noise and parameters that are going to be measured. The rise time of a system is the time in seconds required for the system output to rise from 10 % to 90 % of its final amplitude. The requirement for the system rise time is that it should not be less than one tenth (1/10) of the signal rise time. There is though an exception for sound with extremely short rise time like shock waves. Closely related to the rise time of the system is the slew rate of amplifiers in the used measuring equipment. The slew rate defines the rate at which the output of the system changes over time, often expressed in volts per second. The slew rate of the amplifiers in the measuring system should be high enough to ensure that the no distortion is introduced to the measuring system for the highest frequency of interest. The system overshoot is the maximum amount the output of the system exceeds the real output in response to an input step function. This overshoot should be less than 5 %. The droop of the system is similar to the overshoot but defines the percentage of the amount that the output of the system drops below the idealized value. The requirement to this parameter is 5 % in response to an input step function. The dynamic range of the system is defined as the difference between the un-weighted sound pressure level of the peak of the signal and the un-weighted sound pressure level of the background noise. This dynamic range should at the minimum include a interval from 1 dB above the signals peak sound pressure level to a sound pressure level of minimum 5 dB below the level of interest. The requirements to the bandwidth of measuring system, used for impulse noise, is according to the [DS/ISO 10843, 1998] that the minimum bandwidth shall include the frequency range from the lowest frequency of interest to the highest frequency of interest. If digital integration is used in the process of calculating time integrated parameters the sampling rate of the recording system should be at least three times the highest frequency of interest. This means that if 20 kHz is the highest frequency of interest then at least a sampling rate of 60 kHz should be used. The requirements to the linearity of the system, which should be able to measure the peak signal level without introducing substantial distortion or clipping.
18
2.3. ASSESSMENT OF NOISE
2.3.4
Current Legislation for Impulsive Noise
The current legislation for limits regarding exposure to noise is not defined in the standards but by relevant authorities in the given countries, like the governments or other local authorities dealing with the topic of environment health or noise exposures. In Denmark the allowed dose of noise exposure in a working environment is defined by Arbejdstilsynet (The Danish Working Environment Authority). These noise doses are defined in the announcement of Arbejdstilsynet number 63 of the 6th of February 2006 pursuant to §39, §43, §46, §57, §63, §73 og §84 in the law concerning working environment, refer to consolidation act number 268 of the 18th of March 2005, [Arbejdstilsynet, 2006]. The relevant sections from this document regarding the allowed noise dose in an 8 hour working day are: §10. If the noise exposure exceeds 80 dB(A) or peak values of impulsive noises exceeds 135 dB(C), the employer is only allowed to let the work get carried out if hearing-protective devices are available. §11. No one may be exposed to noise above 85 dB(A) or peak values above 137 dB(C). The values in §10 define the action limits which, if exceeded or reached, demand the employer to provide hearing-protective devices for the employees and inform them about the possible risks of getting hearing damage. The values in §11 define the absolute limit which must not be exceeded. In such a situation immediate action shall be taken to reduce the noise exposure to a value below the limit. Further the cause of the violation of the exceeded limit should be located and technical and organizational precautions should be taken to prevent further violation of the limit. The values have been adapted from the [Directive 2003/10/EC, 2003] conducted by the European Parliament and Council of the European Union. According to the European legislation the national members of the European Union have to implement the directive in the national legislation before 15 February 2006 in order to protect workers across the European Union from the risk of getting hearing damage.
2.3.5
Equal Energy Hypothesis
The Equal Energy Hypothesis (EEH) is a hypothesis that states that an equal amount of energy will cause an equal amount of hearing impairement independent of how it is destributed in time, [NIOSH No. 98-126, 1998]. This means that noise exposure at a certain level and at a certain duration can cause a certain amount of hearing impairment. If the level of the noise is raised but the duration of the exposure is shortened in such a way that the new noise exposure has the same amount of energy as the first noise exposure it should cause the same amount of hearing impairment. Concerning the EEH a doubling of the energy is equal to a increase in level of 3 dB, but in order to keep the total energy exposure constant the time duration should be reduced to the half. This is also known as the 3-dB rule, [Beranek and Ver, 1992, p. 587]. 19
b
CHAPTER 2. IMPULSIVE NOISE
As described in section 2.3.4, the legislation concerning the limits for noise exposure in a working environment is 85 dB(A) over an 8 hour working day. This value is established in order to protect most of the exposed persons from getting hearing damage but mot all. This means that there is only a small probability of getting hearing damage but it is not guaranteed. Looking at this 85 dB(A) in the perspective of the 3-dB rule (EEH), this means that if the level of the noise exposure is raised by 3 dB the exposure time should be halved in order to keep the same probability of hearing damage. This means that one could be exposed to a level of 88 dB(A) for four hours without increasing the risk of getting hearing damage, at least from the equal energy hypothesis point of view. See table 2.2 for a list of the period of time one could be exposed to a certain level. Exposure level [dB(A)]:
85
88
91
94
97
100
103
106
109
Exposure duration :
8h
4h
2h
1h
30 m
15 m
7.5 m
3.75 m
1.875 m
Table 2.2: The effect of the 3-dB rule or equal energy hypothesis related to the exposure time dependent on level. The duration is indicated in hours (h) and minutes (m).
The equal energy hypothesis has been chosen as a foundation for determining the hazardous level of noise exposure. As a descriptor for this equal energy the “sound exposure level” LAE or “equivalent continuous sound pressure level” LAeqT , both A-weighted, is widely implemented in various standards regarding noise assessment. This concept is satisfactory for continuous noise exposures but regarding impulsive noise [Henderson et al., 1990] and [Roberto et al., 1984] has showed that this hypothesis no longer support the prediction of NIHL caused by the energy of this type of noise.
2.4
Summary
The aim of this chapter has been to investigate fundamental theory regarding impulsive noise. First physical properties of impulsive noise has been investigated followed by a analysis of which parameters are most commonly used in literature to describe impulsive noise, which are: peak level, LAeqT , LE , rise time, A-duration, B-duration, kurtosis and crest factor. Psychophysical and physiological aspects of impulsive noise have been investigated in terms of how the human auditory system perceives impulsive signals and especially loudness perception related to impulsive noise. For elaborating the found parameters real impulsive signals should be recorded which will be described in the next chapter. Guidelines for measuring impulsive noise have also been checked in order to perform optimal measurements.
20
C HAPTER Analysis of Measurements
3
According to the aim of this thesis, see section 1, the purpose of this thesis is to investigate whether there is a relationship between changes in parameters and loudness perception. This should be investigated through a subjective listening test. Before executing the listening test it is important to acquire a better understanding of impulse noise both in terms of waveform and in terms of the parameters identified in section 2.1.3. For this purpose recording of impulsive noise should be recorded, primarily for analyzing the parameters and secondly to create a bank of impulse sounds from which the stimuli for the listening test can be chosen or designed. To select or design appropriate sounds files for the listening test the limits of the different parameters should be found. As mentioned in section 2.1.2 impulse noise is a wide area, and a consistent recording and analysis of every type of impulse noise will be more pervasive than the time limit of this project allows. A selection of types of impulsive noise must therefore be found for measuring and analyzing.
3.1
Choosing Signals for Recording
Impulse sounds emitted by explosive sources are difficult to control and recording of them may in most cases by executed in situ. As seen in section 2.1.1 the peak SPL is very large for impulse sounds and rise times are also very short, both sets high demands to the recording equipment. It is therefore chosen to concentrate on impact sounds, which furthermore can be generated in controlled environments, providing that the equipment used for generating the impact sounds are possible to move to the department. The problem with in situ measurements is that there will always be an considerable amount of background noise. As described in section 2.1.2 the most impact noise is emitted in construction sites and in factories. As mentioned church bells and slamming of a door is also frequent impact noises, but typically the peak level of church bells will be of low level and most people will only experience a few doors being slammed everyday so none of them will be especially harmful. In the introduction chapter 1 is described that workers in the metal industry suffers more from hearing impairment than workers in the wood industry when the current legislation limits are conformed. It is therefore chosen to concentrate in impact between wood and metal materials and combinations of these.
21
b
CHAPTER 3. ANALYSIS OF MEASUREMENTS
At the department of acoustics numerous rooms with different room characteristics and reverberation time can be used for the recordings. This gives the opportunity to record the impulse sounds in a controlled environment with a low background noise. Furthermore the recordings of the impact noise will not depend on the working hours of the factory or construction site. For choosing the source generating the noise event it should be noticed that the event should be repeatable, meaning that it should be possible to repeat each sound event and obtain approximately identical waveforms for the recorded signals. Examples of noise sources which could be recorded: • Use of a nail gun. In most nail guns the nail will be accelerated by compressed air, which also emits sound. The material combination is typically metal nails against wood. It is semi-automatic so the sound event should therefore be easy to repeat. • Standard tapping machine is defined in [DS/ISO 140-6, 1998, Annex A]. It is a box with metal weights dropped from a specific height on a base. A standard tapping machine has several weights, which individually can be removed and the dropping frequency can thereby be controlled. It is not typically used in the industry or everyday life. The material combination is metal against every chosen base material. • Hammer on material. It is mostly used in construction. The hammer material can be rubber, metal and wood. The sound event is not easy to repeat since it is difficult to control the force used for every sound event. • One material dropped on another material. Probably not a typical event in the industry but resembles typical noises. The sound event can be set up to be repeatable by controlling the height. The last option is chosen since it can be repeated and gives most options for different controlled setups which also can be varied throughout the recordings.
3.2
Recording of Impact Noise
The impact noise emitting event should be repeatable, which implies that the force of the impact should be controlled together with the direction and duration of the impact. Therefore it is not possible simply to use a hammer and hit something; instead a mechanical device needs to be set up. It is chosen that the event will be a weight falling down on a plate from a specific height controlled by a string as shown at figure B.1 on page 4 in appendix B. For the setups of the recordings it is chosen to vary the following: • Recordings in different environments: – Recordings in an anechoic room to avoid any other sound than the direct sound (referred to as: “anechoic”) – Recordings in an echoic room with both:
22
3.2. RECORDING OF IMPACT NOISE
* A free field microphone placed close to the sound source to avoid reflections from surfaces and thereby an approximate to the direct sound (referred to as: ”near field”) * A diffuse field microphone placed far from the sound source to pick up both the direct sound from the source together with the influence of the room in terms of reflections (referred to as: “diffuse field”) • Materials of plate and weight, four combinations: – Metal weight on metal plate – Metal weight on wood plate – Wood weight on metal plate – Wood weight on wood plate • Setup of plate, see figure B.2 on page 5 in appendix B – The plate placed on rubber feet (referred to as: “Regular” setup) – Regular setup, but suppressed by a cotton bag including pebbles (referred to as: “Suppressed” setup) – The plate upside down (referred to as: “Flipped” setup) • Height of weight above plate for controlling the mechanical energy: – 0.5 m – 1.0 m – 1.5 m – 2.0 m A detailed description of the recordings can be found in appendix B.
Equipment For generating and recording the impulse sounds extended demands for the equipment needs to be set, see section 2.3. The available measuring equipment at the department with the highest samplings frequency is either the 01 dB Symphonie or 01 dB Harmonie acquisition system which both operate with a samplings frequency of 51.2 kHz. For both systems it is easy to calibrate microphones, record signals, find impulse responses and look at signal in one-third octave bands. Computers with a sound card running at a sampling frequency of up to 192 kHz are also available, but it is not as easy to control the calibration of microphones which needs extra equipment, such as measuring amplifier or AGC-unit1 . The controller for the Symphonie and Harmonie systems is also a laptop which will be easy to transport, where the soundcards are in stationary computers. Since it is decided only to record impact noise, the demands are not as tight as if explosions were going to be recorded. Test recordings were performed to check the acquisition system and it was found that the Symphonie system is acceptable for recording the impact sounds. 1 Automatic
Gain Control
23
b 3.3
CHAPTER 3. ANALYSIS OF MEASUREMENTS
Analysis of Recordings
In total 1440 impacts have been recorded divided on three environments, four material combination, three setups and four heights, see previous section. Furthermore each of the recordings has been repeated so ten signals are recorded for each setup. The signals presented in this chapter have all been normalized to have a Leq = 65 dB. The procedure of normalization is described in appendix C. Time plots of the representative2 recordings from respectively the recordings of wood on metal and wood on wood is displayed in figure 3.1 and figure 3.2 to show examples of metal plate and wood plate recordings. Each plot is aligned in time so the peak value is placed at 0 ms, furthermore a black line is drawn illustrating 10 % of the peak value. This line is used both for calculating rise time and B-duration. The parameters described in section 2.1.3 have been calculated for each of the signals. The parameters for the wood on metal recordings are listed in table 3.1 and the parameters for the wood on wood recordings are listed in table 3.2. Environment
Anechoic
Diffuse field
Near field
Setup
Rise Time
Crest Factor
Kurtosis
B-duration
Peak SPL
[ms]
[Pa/Pa]
[ ]
[ms]
[dB] 98.37
Regular
1.52
52.32
572.5
28.5
Suppressed
1.23
44.75
439.2
31.2
96.94
Flipped
1.66
57.35
1195.6
9.6
99.91
Regular
10.31
19.39
66.8
184.8
90.44
9.32
21.00
74.6
178.7
91.23
Flipped
10.18
20.39
77.8
176.8
91.42 97.56
Suppressed Regular
2.29
38.26
251.7
78.0
Suppressed
2.29
34.32
172.5
85.2
97.02
Flipped
2.38
58.75
863.6
36.4
100.42
Table 3.1: Parameters calculated of the wood on metal signals displayed in figure 3.1. Environment
Anechoic
Diffuse field
Near field
Setup
Rise Time
Crest Factor
Kurtosis
B-duration
Peak SPL
[ms]
[Pa/Pa]
[ ]
[ms]
[dB]
Regular
0.18
51.64
677.4
17.6
98.86
Suppressed
0.18
61.51
910.8
20.5
100.47
Flipped
0.31
38.42
578.1
16.9
97.51
Regular
36.58
18.98
91.2
180.1
92.22
Suppressed
18.71
18.42
82.6
182.1
91.69
Flipped
0.33
27.18
82.2
162.7
94.81
Regular
0.94
27.85
224.1
67.6
96.33
Suppressed
0.94
32.90
253.1
67.7
97.83
Flipped
2.38
39.59
325.3
110.7
98.34
Table 3.2: Parameters calculated of the wood on wood signals displayed in figure 3.2.
When comparing the waveforms in the plots in the rows in figure 3.1 and figure 3.2 it is a comparison of the three different environments the signals were recorded in. The first row displays the recordings in the anechoic rooms. Typically a high thin peak value is followed by a very fast decay to a sound pressure barely over the noise floor. For the metal plot it seems that there is a 2A
representative recording is the one of the ten repetitions which is best cross correlated with the mean of the ten repetitions. This is also described in appendix B.
24
Sound Pressure [Pa]
3.3. ANALYSIS OF RECORDINGS
2
2
2
1
1
1
0
0
0
−1
−1 Anechoic Regular
−2
Sound Pressure [Pa]
0
−2
200
0
100 Time [ms]
200
0
2
2
1
1
1
0
0
0
−1
−1
−1
Diffuse field Regular 0
100 Time [ms]
Diffuse field Suppressed
−2
200
0
100 Time [ms]
200
0
2
2
1
1
1
0
0
0
−1 Near field Regular
−2 0
100 Time [ms]
200
100 Time [ms]
200
Diffuse field Flipped
−2
2
−1
Anechoic Flipped
−2
2
−2
Sound Pressure [Pa]
100 Time [ms]
−1 Anechoic Suppressed
100 Time [ms]
200
−1 Near field Suppressed
−2 0
100 Time [ms]
200
Near field Flipped
−2 0
100 Time [ms]
200
Figure 3.1: Time plot of the representative wood on metal recordings for the three different environments (anechoic recordings in the first row, diffuse field recordings in the second row and near field recordings in the third row) and for the three setups (regular setup in the first column, suppressed setup in the second column and flipped setup in the third column), all for a height of 2 m.
25
b
Sound Pressure [Pa]
CHAPTER 3. ANALYSIS OF MEASUREMENTS
2
2
2
1
1
1
0
0
0
−1
−1 Anechoic Regular
−2
Sound Pressure [Pa]
0
−2
200
0
100 Time [ms]
200
0
2
2
1
1
1
0
0
0
−1
−1
−1
Diffuse field Regular 0
100 Time [ms]
Diffuse field Suppressed
−2
200
0
100 Time [ms]
200
0
2
2
1
1
1
0
0
0
−1 Near field Regular
−2 0
100 Time [ms]
200
100 Time [ms]
200
Diffuse field Flipped
−2
2
−1
Anechoic Flipped
−2
2
−2
Sound Pressure [Pa]
100 Time [ms]
−1 Anechoic Suppressed
100 Time [ms]
200
−1 Near field Suppressed
−2 0
100 Time [ms]
200
Near field Flipped
−2 0
100 Time [ms]
200
Figure 3.2: Time plot of the representative wood on wood recordings for the three different environments (anechoic recordings in the first row, diffuse field recordings in the second row and near field recordings in the third row) and for the three setups (regular setup in the first column, suppressed setup in the second column and flipped setup in the third column), all for a height of 2 m.
26
3.4. ANALYSIS OF PARAMETERS
longer ringing. By ringing is meant that the plate is set into vibrations after the impact and thereby continuous to emits sound. This effect is most notable for the regular setup. Besides the ringing there is not much difference between the metal plate recordings and the wood recordings for the anechoic environment. From table 3.1 and table 3.2 it can seen that the anechoic signals typically have a very short rise time, a high crest factor, a high kurtosis a short B-duration and a high peak value. The second row of the plots in figure 3.1 and figure 3.2 displays the recordings obtained with the diffuse field microphone in the echoic room. For the metal plate plots there is a slow rise to the peak value followed by a slow decay. The ringing time is long compared to anechoic. For the wood plate recordings the waveform is different to the metal plate, especially for the regular and suppressed setup. At a time value of approximately -40 ms there is a rise in sound pressure but instead of decaying the sound pressure is approximately constant for 40 ms where the sound pressure rises to the peak value at 0 ms followed by a slow decay. When recording the signals the microphone was positioned so it was ensured that it would also record reflections from the room of the emitted impact sound. From table 3.1 and table 3.2 it can seen that the diffuse field signals typically has long rise time, a small crest factor, a very small kurtosis, a long B-duration and a low peak value. The third row of the plots in figure 3.1 and figure 3.2 displays the recordings executed in the near field of the sound source in the echoic room. Typical for these recordings is a waveform similar to the one obtained in the anechoic room but with a less steep peak and a slower decay. They are approximately a mix of the anechoic signals and the diffuse field signals, which is verified when comparing with the parameters listed in table 3.1 and table 3.2. The parameter values read on the tables for the near field signals are typically between the values for the anechoic signals and the diffuse field signals. For all environments the flipped setup gives typically a high narrow peak value which decays steeply until a sound pressure of approximately 10 % of the peak value. After this point the following decay is slow. This behavior is most notable for the anechoic signals and the near field signals. In table 3.1 and table 3.2 the flipped setup seems not to give any typical behavior when comparing them to the parameters for the regular and suppressed setup. The waveform of the regular and suppressed setup seems to be quite similar, with a narrow peak which rapidly decays in a smoother curve than for the flipped setup. No typical behavior can be found for the values in table 3.1 and table 3.2 except that the values from the regular and suppressed setup are quite similar compared to the values from the flipped setup.
3.4
Analysis of Parameters
In section 2.1.3 a list of parameters to describe impulse noise are identified including the existing parameter for evaluating noise, LAeq .
3.4.1
LXeq
As mentioned in the introduction in chapter 1 the amount of hearing damage experienced by wood workers has been lower than the amount of hearing damage experienced by metal workers when
27
b
CHAPTER 3. ANALYSIS OF MEASUREMENTS
they have been exposed to the same equivalent level of noise, LAeq . Working with metal material typically emits higher frequencies than working with wood material which suggests that the metal workers have been exposed to noise with higher frequency content than wood workers. A-weighting seems not to be the correct weighting to apply and the C-weighting and no weighting should also be tested. Each of the parameters under test should be calculated on signals which are normalized to have either the same Leq , LAeq or LCeq . A consistent analysis of each parameter for all three weightings is quite extensive and they will therefore only be analyzed for signals normalized to have the same Leq . Leq is chosen because it is un-weighted, which means that the parameters will be compared for signals with their physical values and not normalized to a subjective level. In the following sections each parameter except the A-duration and LE will be calculated for each of the normalized impact noise recordings and thereby evaluated. The signals are normalized to a Leq value of 65 dB. The normalization procedure is described in appendix C. The A-duration is not included since it is judged to be most relevant for explosive sound sources. The LE is not included since it basically express the same as Leq since both are descriptors of energy integrated over a certain time period and if this period is equal the same relationship will be expressed. LXeq will be used as the common descriptor for all three weightings.
3.4.2
Rise Time
The rise time is defined as the time difference between 10 % and 90 % of the peak sound pressure level. Figure 3.3 shows a histogram of the rise time for the 1440 recordings distributed for the three types of recording; the recording in the anechoic room and the two different microphones used for the recordings in the echoic room. In the figure the distribution of the different rise times is displayed. For the anechoic room most of the rise times are approximately 0.2-0.3 ms which is equal to approximately 10 samples with a samplings frequency of 51.2 kHz. Furthermore, nearly all rise times are less than 2 ms. A similar situation can be found for the near field microphone in the echoic room where most rise times are located around 1 ms and between 2 ms and 3 ms. For the diffuse field microphone in the echoic room the distribution of the rise time is very different compared to the other two cases. Here the majority of the rise times are distributed at 1 ms but then a group of rise time is spread up to a rise time as high as 18 ms, and furthermore there is a distribution between 35 ms and 40 ms. Diffuse Field Microphone
100
50
0
0
10
20 30 Rise time [ms]
40
Near Field Microphone Numbers of occurence of rise times
Numbers of occurence of rise times
Numbers of occurence of rise times
Anechoic Room 150
150
100
50
0
0
10
20 30 Rise time [ms]
40
150
100
50
0
0
10
20 30 Rise time [ms]
40
Figure 3.3: Histogram for the rise times of the 1440 recordings for each of the three types of recording; the recording in the anechoic room and the two different microphones used for the recordings in the echoic room.
In figure 3.4 the 1440 recordings are divided related to the different environment, setups and heights they were recorded in. Based on figure it is possible to investigate the cause of the distributions from figure 3.3. For the diffuse field microphone the rise times distributed between 35 ms 28
3.4. ANALYSIS OF PARAMETERS
anechoic room rise time [ms]
and 40 ms is mainly due to the metal on wood and wood on wood recordings, and only for the two plate setups: regular and supressed. 40
40
40
30
30
30
20
20
20
10
10
10
diffuse field mic. rise time [ms]
0
0
0
h=0.5 h=1.0 h=1.5 h=2.0
40
40
40
30
30
30
20
20
20
10
10
10
0
near field mic. rise time [ms]
h=0.5 h=1.0 h=1.5 h=2.0
h=0.5 h=1.0 h=1.5 h=2.0
0
0
h=0.5 h=1.0 h=1.5 h=2.0
40
40
40
30
30
30
20
20
20
10
10
10
0
0
0
h=0.5 h=1.0 h=1.5 h=2.0 Repetitions of combinations of height and regular setup metal plate metal weight
h=0.5 h=1.0 h=1.5 h=2.0 Repetitions of combinations of height and suppressed setup
wood plate metal weight
metal plate wood weight
h=0.5 h=1.0 h=1.5 h=2.0
h=0.5 h=1.0 h=1.5 h=2.0
h=0.5 h=1.0 h=1.5 h=2.0 Repetitions of combinations of height and flipped setup
wood plate wood weight
Figure 3.4: Rise times for all 1440 recordings. The three rows displays respectively the recordings from the anechoic room, the recording from the diffuse field microphone and the recording from the free field microphone, both in the echoic room. In the plots in the columns is respectively the plate setup types which is again divided in the four heights. The four different colors display the four combinations of the material.
In figure 3.5 a time plot of a recording from the diffuse field recording is shown. The combination of material is wood on metal, the plate is suppressed and the height is 2 m. As usually there is a large increase in sound pressure but instead of decaying rapidly it only decreases slightly. This is followed by a second rise to the peak value followed by a slow decay. The peak value should have been the peak marked with a circle, but instead a larger peak arises after approximately 40 ms. The same is the case for a huge part of the recordings of the diffuse field microphone. Most rise time has a value less than 5 ms (256 samples). If rise time is to be used as a parameter for the listening test, stimuli has to be created where the rise time is changed between 0 ms and 5 ms. The differences will be so small that it probably not will be perceived by the human ear. Furthermore a parameter with this small differences in values are very vulnerable to noise and will also demand some really good equipment in order to reproduce this fast change.
29
b
CHAPTER 3. ANALYSIS OF MEASUREMENTS
Sound Pressure [Pa]
0.6 0.4 0.2 0 −0.2 −0.4 −0.6 −50
0
50 Time [ms]
100
150
Figure 3.5: Time plot of a diffuse field supressed setup recording. The horizontal solid lines marks 10 % and 90 % of the peak value. The peak value is at time 0 ms. The circle mark a peak just below the 90 % limit.
3.4.3
B-duration
The B-duration is calculated as the time difference between the first and last exceedence of 10 % of the peak value by the absolute sound pressure, see appendix A. The distribution of this difference for the recorded signals is investigated in figure 3.6 for the three environments. For the signals recorded in the anechoic room the B-duration is mostly shorter than 100 ms, with the majority centered around a B-duration of 10 ms. This is as expected since most signals in the anechoic room decay very quickly. For the diffuse field recordings the majority is distributed around approximately 180 ms but with a wide spread between 110 ms and 310 ms. This is also as expected since reflections from the room of the impact noise were also recorded. A long decay time is therefore expected and thereby also a long B-duration. For the near field signals there is no distinct majority while the values are mostly concentrated in values from 40 ms to 160 ms but also with a small group above 200 ms.
30 20 10 0
0
100 200 B−duration [ms]
300
Numbers of occurence of B−duration
Numbers of occurence of B−duration
40
Near Field Microphone
50 40 30 20 10 0
0
100 200 B−duration [ms]
300
Numbers of occurence of B−duration
Diffuse Field Microphone
Anechoic Room 50
50 40 30 20 10 0
0
100 200 B−duration [ms]
300
Figure 3.6: Histogram for the B-durations of the 1440 recordings for each of the three types of recording; the recording in the anechoic room and the two different microphones used for the recordings in the echoic room.
These relationships are more closely investigated in figure 3.7 where the B-duration values are displayed for each of the recorded signals. Generally it seems that all wood plate recordings deviate less than the metal plate. A special case is with the near field recordings in the flipped setup, it seems that either the signals have a value of 170 ms or 120 ms. It can be used as stimuli for the listening test since generally the deviation in B-duration is large for some parameters and approximately constant for other parameters for both metal and wood plate recordings. It is judged that the deviation for some setups is large enough that three levels of the parameters can be found. For example for the green dots (metal weight, wood weight) in the first plot in the top row in figure 3.7 the spread in B-duration is large enough that both a signals with a B-duration of 40 ms, a signal with a B-duration of 80 ms and a B-duration of 120 ms and thereby three levels of B-duration.
30
anechoic room B−duration [ms]
3.4. ANALYSIS OF PARAMETERS
300
300
300
200
200
200
100
100
100
diffuse field mic. B−duration [ms]
0
0
0
h=0.5 h=1.0 h=1.5 h=2.0
300
300
300
200
200
200
100
100
100
0
near field mic. B−duration [ms]
h=0.5 h=1.0 h=1.5 h=2.0
h=0.5 h=1.0 h=1.5 h=2.0
0
0
h=0.5 h=1.0 h=1.5 h=2.0
300
300
300
200
200
200
100
100
100
0
0
0
h=0.5 h=1.0 h=1.5 h=2.0 Repetitions of combinations of height and regular setup metal plate metal weight
h=0.5 h=1.0 h=1.5 h=2.0 Repetitions of combinations of height and suppressed setup
wood plate metal weight
metal plate wood weight
h=0.5 h=1.0 h=1.5 h=2.0
h=0.5 h=1.0 h=1.5 h=2.0
h=0.5 h=1.0 h=1.5 h=2.0 Repetitions of combinations of height and flipped setup
wood plate wood weight
Figure 3.7: B-duration for all 1440 recordings. The three rows displays respectively the recordings from the anechoic room, the recording from the diffuse field microphone and the recording from the free field microphone, both in the echoic room. In the plots in the columns is respectively the plate setup types which is again divided in the four heights. The four different colors display the four combinations of the material.
3.4.4
Crest Factor/Peak Value
The parameter crest factor is defined as the ratio between the peak sound pressure level and the RMS value of the signals, see section 2.1.3. Since all parameters in this section are calculated on signals which all have the same Leq it means that all signals analyzed in this section also have identical RMS values. The only difference between the crest factor and the peak value will therefore be a constant factor. It is chosen to display the values of the peak value in this section since it is easier to understand the meaning of a peak value since it is more intuitive. The distribution of the peak value for the recorded signals is displayed in the histogram in figure 3.8 for the three environments. For the anechoic room the values are distributed equally from approximately 95 to 104 Pa, while for the diffuse field recordings, the values are spread in a smaller area from approximately 88 to 95 Pa. For the anechoic and the diffuse field signals the distribution of the crest factor are approximately normal distributed, but for the near field signals the distribution is different. Most of the values are distributed from 92 to 97 Pa, but a small amount is distributed between 97 and 101 Pa. Resultantly there are peak values of impulse signals distributed from 88 Pa to 101 Pa even though all signals are normalized to have the same equivalent level.
31
b
CHAPTER 3. ANALYSIS OF MEASUREMENTS
5
0
90
95 100 Peak Value [dB]
Numbers of occurence of peak value
Numbers of occurence of peak value
10
15
10
5
0
90
95 100 Peak Value [dB]
Near Field Microphone Numbers of occurence of peak value
Diffuse Field Microphone
Anechoic Room 15
15
10
5
0
90
95 100 Peak Value [dB]
Figure 3.8: Histogram for the peak value of the 1440 recordings for each of the three types of recording; the recording in the anechoic room and the two different microphones used for the recordings in the echoic room.
In figure 3.9 the distribution is displayed for all combinations of environments, setups, heights and materials. Overall it can be seen that, within each environment for the regular and the suppressed setup, the peak values are quite similar where the signals with the flipped setup stand out for most combinations. Peak value can also be used as a parameter for the listening test. The recorded signals are also usable as test stimuli even that the difference in peak level is not huge, and only two levels of comparison are available.
3.4.5
Kurtosis
The kurtosis parameter expresses the relationship between the amounts of samples close to the peak value compared to the amount of samples close to the mean value. Thereby the kurtosis to some degree expresses the same as the crest factor. In figure 3.10 is seen that the deviation of the kurtosis values are huge for the signals. The signals recorded in the anechoic room are spread equally over a range of kurtosis from 200 to 1700, where the kurtosis for the diffuse field recordings is distributed from 0 to 200. This huge difference is caused by the fact that the peak is very narrow for the anechoic recordings, where the peak for the diffuse field recordings is broader. The majority of the kurtosis values for the near field recordings are located between 0 and 400 centered on 200. A small group is located between 500 and 1000. In figure 3.11 it can be seen that this is due to the metal plate recordings in the flipped setup. It is complicated to compare the kurtosis values for the three environments since the difference in their values are pronounced. The greatest variance is observed with the flipped setup, where the wood plate signals seem to be unaffected in the near field recording but hugely deviates for the anechoic recording. For the flipped setup with the anechoic recording it should also be noted that there are a pronounced difference in the wood and metal weight recordings for the wood plate. Concerning this parameter there will be a basis for choosing test files. Generally it can also be noticed that the kurtosis values tend to be opposite to the B-duration values.
32
anechoic room peak value [dB]
3.4. ANALYSIS OF PARAMETERS
100
100
100
95
95
95
90
90
90
diffuse field mic. peak value [dB]
h=0.5 h=1.0 h=1.5 h=2.0
h=0.5 h=1.0 h=1.5 h=2.0
100
100
100
95
95
95
90
90
90
h=0.5 h=1.0 h=1.5 h=2.0
near field mic. peak value [dB]
h=0.5 h=1.0 h=1.5 h=2.0
h=0.5 h=1.0 h=1.5 h=2.0
h=0.5 h=1.0 h=1.5 h=2.0
100
100
100
95
95
95
90
90
90
h=0.5 h=1.0 h=1.5 h=2.0 Repetitions of combinations of height and regular setup
h=0.5 h=1.0 h=1.5 h=2.0 Repetitions of combinations of height and suppressed setup
metal plate metal weight
wood plate metal weight
metal plate wood weight
h=0.5 h=1.0 h=1.5 h=2.0 Repetitions of combinations of height and flipped setup
wood plate wood weight
Figure 3.9: Peak value for all 1440 recordings. The three rows displays respectively the recordings from the anechoic room, the recording from the diffuse field microphone and the recording from the free field microphone, both in the echoic room. The plots in the columns show respectively the plate setup types which again is divided in the four heights. The four different colors display the four combinations of the material. Diffuse Field Microphone
80 60 40 20 0
500
1000 Kurtosis
1500
100 80 60 40 20 0
Near Field Microphone Numbers of occurence of kurtosis
Numbers of occurence of kurtosis
Numbers of occurence of kurtosis
Anechoic Room 100
500
1000 Kurtosis
1500
100 80 60 40 20 0
500
1000 Kurtosis
1500
Figure 3.10: Histogram for the kurtosis values of the 1440 recordings for each of the three types of recording; the recording in the anechoic room and the two different microphones used for the recordings in the echoic room.
3.4.6
Frequency Content
This section will contain description of the frequency contents of the recorded files. This will be done by time frequency plots calculated by the MATLAB function SPECGRAM which uses the discrete short time Fourier transform. For the displayed time frequency plots is chosen to normalize each signals to the peak value. This results in that the peak value will be indicated by black. This
33
b anechoic room kurtosis
CHAPTER 3. ANALYSIS OF MEASUREMENTS
1500
1500
1500
1000
1000
1000
500
500
500
diffuse field mic. kurtosis
h=0.5 h=1.0 h=1.5 h=2.0
h=0.5 h=1.0 h=1.5 h=2.0
1500
1500
1500
1000
1000
1000
500
500
500
h=0.5 h=1.0 h=1.5 h=2.0
near field mic. kurtosis
h=0.5 h=1.0 h=1.5 h=2.0
h=0.5 h=1.0 h=1.5 h=2.0
h=0.5 h=1.0 h=1.5 h=2.0
1500
1500
1500
1000
1000
1000
500
500
500
h=0.5 h=1.0 h=1.5 h=2.0 Repetitions of combinations of height and regular setup metal plate metal weight
h=0.5 h=1.0 h=1.5 h=2.0 Repetitions of combinations of height and suppressed setup wood plate metal weight
metal plate wood weight
h=0.5 h=1.0 h=1.5 h=2.0 Repetitions of combinations of height and flipped setup
wood plate wood weight
Figure 3.11: Kurtosis values for all 1440 recordings. The three rows displays respectively the recordings from the anechoic room, the recording from the diffuse field microphone and the recording from the free field microphone, both in the echoic room. In the plots in the columns is respectively the plate setup types which is again divided in the four heights. The four different colors display the four combinations of the material.
is done in order to ease comparison of the different time frequency plots. In the following both the term: “time frequency plot” and “spectrograms” will be used to refer to these kind of plots. As usual the peak value will also be aligned in time to be at 0 ms. As seen in the previous sections the two set of metal plate recordings (both with wood and metal weight) often have similar parameters. In the same way the two sets of wood plate recordings also often have similar parameters. It could therefore be interesting to investigate whether the two sets of data can be put together. In figure 3.12 is shown the time frequency plot of each of the four material combinations for the representative recordings in the anechoic room with the regular setup and a height of 2 m. On these plots it is chosen to display values from the peak value and 70 dB down. The metal plate recordings are displayed in the top row of the plots, where spectrograms of the wood plate are displayed in the bottom row of the plots. It can be seen that the metal plate spectrograms are very similar and that the wood plate spectrograms are very similar. The only noticeable difference is at a frequency of 11.5 kHz where for both metal weight spectrograms (the first column of the plots) there are a distinct artifact lasting approximately 200 ms with a value of less than -50 dB of the peak value. Additional signals were checked and this ringing was present for the majority of them. This is most notable for the anechoic recordings and is created when the metal weight hits the plate it is set into vibration which causes the ringing. The influence from this ringing will most likely be very small but it is even so chosen to concentrate only on the
34
3.4. ANALYSIS OF PARAMETERS
wood weight recordings. When in the remaining part of the report is referred to metal or wood recordings it respectively means the wood on metal recordings and the wood on wood recordings. Wood on metal 20
15
15
Frequency [kHz]
Frequency [kHz]
Metal on metal 20
10
5
0
100
200
10
5
300
0
20
15
15
10
5
0
100 Time [ms]
200
300
Wood on wood
20
Frequency [kHz]
Frequency [kHz]
Metal on wood
100
200
10
5
300
0
100 Time [ms]
200
300
Colorbar
−70
−60
−50
−40
−30
−20
−10
0
Figure 3.12: Time frequency plot of each of the four material combinations for the representative recordings in the anechoic room with the regular setup and a height of 2 m.
The following description of the time frequency plots will be applicable for the time frequency figures used in the rest of the report. Each plot is normalized to so their peak value has a value of 0 dB which according to the colorbar gives a black color in the time frequency plot. The lower limit for the plots are set to -20 dB equivalent to that the topmost 90 % of the signal is displayed. Both rise time and B-durations is calculated for the topmost 90 % of the signals which is also judged to be the most important. In the time axis the signals are aligned in time so the peak value will be displayed at 0 ms. The majority of the signals has no information above -20 dB before -50 ms and after 250 ms which will be used as the limits on the time axis. In the following time frequency plots are analyzed of the same signals shown in the time plots in figure 3.1 on page 25 and figure 3.2 on page 26, which is a time plots of the representative signals recorded in a height of 2 m and for each of the three environments and for each of the three plate setups. Figure 3.13 displays the metal plate recordings and figure 3.14 displays the wood plate recordings. In the frequency axis the frequencies between 20 Hz and 6 kHz are displayed, since these are the frequencies judged to contain the most valuable information. Generally it can be seen that for the metal plate recordings the frequency contents is spread from approximately 100 Hz to 6 kHz whereas the frequency contents for the wood plate typically are 35
b
CHAPTER 3. ANALYSIS OF MEASUREMENTS
10
10 Anechoic Regular
Frequency [kHz]
8
6
4
4
4
2
2
2
100 Time [ms]
200
0
100 Time [ms]
200
10 Diffuse field Regular
8
0
Diffuse field Suppressed
8
6
4
4
4
2
2
2
200
10
0
100 Time [ms]
200
10 Near field Regular
8
0
Near field Suppressed
8
6
4
4
4
2
2
2
200
200
0
100 Time [ms]
200
Near field Flipped
8
6
100 Time [ms]
100 Time [ms]
10
6
0
200
Diffuse field Flipped
8
6
100 Time [ms]
100 Time [ms]
10
6
0
Anechoic Flipped
8
6
10
Frequency [kHz]
Anechoic Suppressed
8
6
0
Frequency [kHz]
10
0
100 Time [ms]
200
Colorbar
Figure 3.13: Time frequency plots of the representative wood on metal recordings for the three different environments (anechoic recordings in the first row, diffuse field recordings in the second row and near field recordings in the third row) and for the three setups (regular setup in the first column, suppressed setup in the second column and flipped setup in the third column), all for a height of 2 m.
36
3.4. ANALYSIS OF PARAMETERS
10
10 Anechoic Regular
Frequency [kHz]
8
6
4
4
4
2
2
2
100 Time [ms]
200
0
100 Time [ms]
200
10 Diffuse field Regular
8
0
Diffuse field Suppressed
8
6
4
4
4
2
2
2
200
10
0
100 Time [ms]
200
10 Near field Regular
8
0
Near field Suppressed
8
6
4
4
4
2
2
2
200
200
0
100 Time [ms]
200
Near field Flipped
8
6
100 Time [ms]
100 Time [ms]
10
6
0
200
Diffuse field Flipped
8
6
100 Time [ms]
100 Time [ms]
10
6
0
Anechoic Flipped
8
6
10
Frequency [kHz]
Anechoic Suppressed
8
6
0
Frequency [kHz]
10
0
100 Time [ms]
200
Colorbar
Figure 3.14: Time frequency plots of the representative wood on wood recordings for the three different environments (anechoic recordings in the first row, diffuse field recordings in the second row and near field recordings in the third row) and for the three setups (regular setup in the first column, suppressed setup in the second column and flipped setup in the third column), all for a height of 2 m.
37
b
CHAPTER 3. ANALYSIS OF MEASUREMENTS
concentrated in the low frequencies particularly from approximately 100 Hz to 2 kHz. This is however not the case for the regular and suppressed plate setup in the anechoic room where the frequency response is more wide spread than the corresponding time frequency plots for the metal plate. The time frequency plots for the recordings from the anechoic room shows a wide spread in frequency at 0 ms followed by a fast decay, as described for the time plots of the recordings from the anechoic room. For the diffuse field recordings the slower decay is clearly visible. For the recordings of the metal plate placed on the feet the prominent frequencies are located between 3 kHz and 5 kHz starting at time -10 ms. At time 0 ms (the time for the peak value) the frequency contents located at approximately 400 Hz starts to have an influence. For the near field recordings the frequency content are the same but starting at time 0 ms. Since the main frequencies are located from 3 kHz to 5 kHz for the anechoic recordings with the plate on the feet a conclusion could be that main frequencies for the metal plate on the feet are located between 3 kHz and 5 kHz and that the influence of the rooms amplifies the frequencies at 400 Hz. The frequency contents for the flipped setup is concentrated below 2 kHz. For the wood plate recordings on the feet most frequency contents are located at approximately 400 Hz which above is described as room influence. For the flipped recordings the frequency responses are quite similar to the frequency response from the metal recordings in the flipped setup. Generally the frequency content is usable as a parameter since the frequency contents between the wood plate and the metal plate are quite different. Within each material combination signals with approximately constant frequency contents can also be found.
3.5
Choice of Parameter
Based on the previous section following conclusions can be summarized for each parameter: • Rise Time – Not usable as a parameter for the listening test since the deviations in the parameter is so small that it will probably not be perceived, and furthermore such small deviations are very sensitive to the changes created by the reproduction system. By changing the files the rise time can be made longer, but since there is no information of perceptual limits a pilot test should be executed to create realistic stimuli. • B-duration – Usable as a parameter, since it deviates for some setups and is approximately constant for other setups. The deviation is large enough to create three levels of comparison if the recorded signals are used as stimuli for the listening test. • Peak value – Usable as a parameter, are also deviating for some setups and approximately constant for other setups. The deviation is not large and the comparison should be performed in two levels if the recorded signals are used as stimuli for the listening test. • Crest factor
38
3.5. CHOICE OF PARAMETER
– Usable as a parameter but for the Leq normalized signals the crest factor is identical to the peak value except for a scaling in amplitude. For LAeq and LCeq there will be some difference in crest factor. It is basis for two levels of comparison if the recorded signals are used as stimuli for the listening test. • Kurtosis – Usable as a parameter but expresses approximately the same relationship as the crest factor. The spread in kurtosis for the three environments are very different and can be basis for more levels than three. • Frequency content – Usable as a parameter. Since the wood plate and metal plate generally have quite different frequency characteristics there should be basis for test files. Signals with approximately constant frequency contents can also be found within each material combination. • LXeq – Should not be varied in the test, but the influence should be tested by holding it constant for each of the following weightings: un-weighted, A-weighted and C-weighted. Another method for comparing the five first parameters (Rise time, B-duration, Peak value, Crest factor and Kurtosis) could be a cross correlation of each combination of parameters. The frequency contents are not included since it is difficult to describe the frequency contents with a single value. For each parameter a vector is created with the values of the parameter calculated for all 1440 signals sorted in the same way. When the MATLAB command: max(abs(xcorr(x,y,’coeff’))) are used on the vectors, where x and y denotes the vectors containing the two compared parameters a single value between zero and one is the result. This value tells how correlated the two vectors are, disregarding delay, amplitude scaling and reversal of sign. The closer the value is towards zero the less the vectors are correlated and the closer the value is towards one, the more correlated the vectors are, meaning that the two compared parameters will basically provide the same information about the signals. The maximum absolute cross correlation between the five vectors is shown in figure 3.15. Maximum absolute cross correlation of the five parametres 1
Kurtosis
0.9
0.8
Crest Factor
0.7
0.6
Peak Value
0.5
0.4
B−duration
0.3
0.2
Rise Time
0.1
Rise Time
B−duration
Peak Value
Crest Factor
Kurtosis
0
Figure 3.15: Maximum absolute cross correlation of the five first parameters. The darker the color the more correlated the two compared parameters are.
39
b
CHAPTER 3. ANALYSIS OF MEASUREMENTS
As mentioned above the crest factor and peak value are identical except for scaling, this is confirmed by figure 3.15 where the maximum cross correlation of them yields a value of one, which means they are identical. It is also mentioned that the crest factor and the kurtosis express similar relations in the signals which is confirmed by the figure where the maximum cross correlation of them has a value of 0.9. For the rise time and the B-duration there seem to be little correlation with the other parameters. Based on the conclusions in the itemize and figure 3.15 it is chosen to use the following parameters for the listening test: • B-duration • Peak level • Frequency content These parameters are also chosen since they describe three different aspects of the signals. The B-duration is highly related to the time aspect, especially of the decay of the signals. Peak level is more related to level and finally the frequency content is related to the frequency domain of the signals.
3.6
Listening Test Stimuli
Based on section 3.3 to 3.5 it can be concluded that the recorded sound files should be a sufficient base of stimuli for the listening test. If necessary the signals can be manipulated if they are not entirely suitableas test stimuli, but care should be taken and recalculation of the parameters should be executed if the signals are modified. In the previous section the three parameters B-duration, peak level and frequency contents are chosen as the usable parameters for the test. Simultaneously the three frequency weightings should be evaluated: un-weighted, A-weighted and C-weighted. This leads to the following three test hypothesis, the first three under the condition that the compared sounds all has the same LXeq value: 1. A difference in B-duration is related to a perceived difference in loudness 2. A difference in peak value is related to a perceived difference in loudness 3. A difference in frequency content is related to a perceived difference in loudness 4. There is no perceived difference in loudness for signals with the same normalizations, LAeq , LCeq or Leq . This should be tested by comparing two impact sounds in terms of perceived loudness. For the two compared sounds only one of the three parameters should be varied while the other two parameters should be kept constant. E.g. for the test part where the B-duration is varied both the peak value and the frequency content should be kept approximately constant.
40
3.6. LISTENING TEST STIMULI
The test should be divided into three parts, one for each parameter, and the sets of test stimuli used for each part should consist of an equal amount of signals from each normalization to LXeq , The parameters can only be kept constant to a certain degree, e.g. it can be difficult to find enough stimuli with identical B-duration and identical peak value but a huge difference in frequency contents. Since there are no available information regarding the smallest perceivable difference for impact sounds for the three chosen parameters the limits must be decided. These limits will also be evaluated as a side benefit in the listening test. Based on section 3.5 the following limits are chosen under the assumption that a difference in loudness will not be perceived if the difference in the parameters is smaller than the limits: • B-duration: 10 ms • Peak value: 1 dB • Frequency content: Based on the same environment and setup. Further by visual inspections of time frequency plots of the first 20 ms following the peak value it should be judged that the frequency contents are fairly identical followed by an aural inspection where no distinct frequency contents should be present. These limits will be used when the test stimuli are selected. From section 3.6 the possible numbers of levels for each parameter were also described to be three for the B-duration and two levels for both the peak level and the frequency contents. This means that three compared stimuli should be used for the test of the B-duration and two compared stimuli for the test of the peak level and frequency content.
41
C HAPTER Listening Test
4
This chapter will describe all matters concerning the listening test: test method to apply is analyzed, test stimuli is found and the test is set up including setup of equipment. In the analysis 3.6 four hypothesis are set in order to evaluate the chosen parameters as described in the aim of the thesis in chapter 1. The first three parameters are under the condition that the compared sound all has the same LXeq value: 1. A difference in B-duration is related to a perceived difference in loudness 2. A difference in peak value is related to a perceived difference in loudness 3. A difference in frequency content is related to a perceived difference in loudness 4. There is no perceived difference in loudness for signals with the same normalizations, LAeq , LCeq or Leq . These four hypothesis will be tested through a subjective listening test.
4.1
Methods for Testing the Hypothesis
The method used for the listening test shall be suited for test of discrimination between two successively presented stimuli. The main topic of the test can be described as loudness discrimination between two successively presented impulses with small equal energy (LXeq ) and small physysical differences. It is in this thesis not intended to sample the whole psychometric curve in order to find the just noticeable difference (jnd) of loudness of impulsive signals for variations in the three different parameters described above. The purpose is rather to investigate whether a change in the mentioned parameters can be related to a change in loudness perception or not.
43
b 4.1.1
CHAPTER 4. LISTENING TEST
Method of Constant Stimuli
The method of constant stimuli is chosen as the test method used in the listening test since it, according to [Snodgrass et al., 1985], is a suitable test method for a discrimination test. The listening test is based on this method even though only a perceived difference in loudness is going to be the test topic rather than sampling the entire psychometric function. The test is going to be a comparison of two successively presented stimuli. Stimuli A and stimuli B, the subjects task will then be to choose the stimuli they perceived as the loudest. When the subjects are incapable of perceiving a difference in terms of loudness they will have to guess which stimuli is the loudest. This is known as a 2 interval forced choose procedure (2IFC).
4.2
Design of Listening Test
In section 3.6 it is described that there should be three test parts, one for each parameter. For each test part there should be an equal amount of files normalized to each weighting. The number of possible levels for each parameter was also identified to be three for the B-duration and two for both the peak level and the frequency content. The total number of comparisons in the test should not be to large since it will tire the subjects who then will be unconcentrated and might answer inconsistently. It is judged to be more important to have concentrated test subjects and therefore a shorter test with fewer comparisons and then maybe more subjects, rather than a too long test with few subjects. By use of pilottest the adequate number of comparison was found for each parameter to be 72 for the B-duration and 48 for both the peak level and frequency content. Table 4.1 presents an overview of the allocation of the files, number of sets for each type of normalization and number of files in each set. Part of test B-duration Peak Value Frequency Content
LXeq
Env./Mat.
# of sets
# of files in each set
A,C,U
Metal
1
3
A,C,U
Wood
1
3
A,C,U
Metal
2
2
A,C,U
Wood
2
2
A,C,U
Diffuse Field
2
2
A,C,U
Near Field
2
2
Table 4.1: Overview of the different parts of the listening test. The values in column “# of sets” and column “# of files in each set” are for each type of energy normalization.
Catch trials, with a obvious difference in the loudness sensation, should be implemented in the test for two reasons. First to be able to check the reliability of the subjects, meaning that the subject should not be in doubt about which stimuli to choose but chose the obvious louder stimuli for all catch trials. The second reason for including the catch trial is to motivate the subjects, by presenting them with pairs of stimuli where they are clearly able to perceive a difference in loudness, and thereby support the subjects in the believe that they are doing the right thing and not just guessing all the time. One set of catch trial is found for each set of stimuli for the B-duration test part and for half of the sets of stimuli for the peak value test part and the frequency content test part.
44
4.2. DESIGN OF LISTENING TEST
4.2.1
Number of Repetitions
If the stimuli for the test are too identical for the test subjects to tell which one is loudest, they will randomly choose one of the presented samples. A way to minimize this guessing is to use the 2IFC procedure and repeat each combination of stimuli four times. If the test subjects then respond the same at least three times out of the four presented (75 %) it is an indication of that the test subject can tell the difference between the loudness generated by the presented pair of stimuli. Each combination of stimuli will therefore be presented four times. In order to minimize the influence caused by the presented order of the stimuli it has been chosen to balance the presentations of the stimuli. This means that each pair of stimuli (A and B) will be presented two times in the order A-B and two times in the opposite order B-A, four times in total. In this way, if the order has any influence on the perceived loudness, both stimuli will be presented as the first two times and the effect of the presented order is neglected.
4.2.2
Presentation of Stimuli
The pairs of compared stimuli are presented in random order which means that that the subjects are not able to predict a sequence in the presentation of the stimuli. For the test for B-duration and for peak value both compared sounds are either based on the metal plate or on the wood plate. One method could be to first present all metal plate stimuli and then all wood plate stimuli, another method could be to mix them. If all comparisons with one type of plate is presented as the first and the other plate comparisons are presented next, it could happen that the test subject can tune in on the differences. By tune in is meant that they can learn what to listen for and by this more sharply distinguish between the presented samples. This is unwanted since then there will be difference in how the first comparisons are perceived compared to the last comparisons in the test part which means that there could be a learning process. Instead if the presentation is mixed so the pair of impulse sound to compare are randomly chosen between metal plate and wood plate recordings the test subject will not in the same way learn to tune in on the differences. It is not possible to completely avoid a learning process, and the only way to compensate for this is to randomize or balance the order of presentation of the stimuli. The catch trials should also be presented randomized between the other samples but for the same test part. There are six sets of catch trials consisting of one pair of stimuli for each test part. The appearance of them should be random, so the test subject does not get suspicious. The test parts will be organized as follows: A list is constructed with all pairs to be compared and to this list the catch trials should be added. For each test subject the presentation order of this list will be randomized.
45
b 4.2.3
CHAPTER 4. LISTENING TEST
Order of Test Parts
For the three parts of the test the learning process can also have an influence. Therefore are the order of the presentation of the test parts balanced, so each test part is presented first, second and last an equal amount of times. (This is not completely true since the number of subjects that participated (19) is not dividable by the number of combinations of the test parts (6)). The test orders and their number of occurrence are listed in table 4.2. 1st Test Part
2nd Test Part
3rd Test Part
Frequency Content
Peak Value
B-duration
4
Frequency Content
B-duration
Peak Value
3
Peak Value
Frequency Content
B-duration
3
Peak Value
B-duration
Frequency Content
3
B-duration
Peak Value
Frequency Content
3
B-duration
Frequency Content
Peak Value
3
# of Occurrence
Table 4.2: Overview of the presentation order of the three test parts in the listening test.
4.3
Test Subjects
A high number of test subjects leads theoretically to a small deviation in the total response, meaning that if the answers from a few test subjects deviates much from the mean the total response will be unaffected if the number of test subjects is large. But executing a listening test also involves a lot of time spent on each test subject, so the more test subjects the more hours are spent. It is therefore judged that a number of subjects between 12 and 18 should be sufficient for this thesis. The test subjects chosen for this listening test should be considered to have a normal hearing. By normal hearing is defined that they do not have any hearing problem meaning a maximum hearing level of 20 dBHL and a normal functioning middle ear. This should be checked as a part of the listening test by respectively an audiometry, a tympanometry and an acoustic reflex test. Additionally the test subjects should complete a questionnaire regarding their history of hearing. The hearing check is described in appendix E.1.1. To avoid translations of test instruction and graphical user interface these are executed in English and only test subjects which understand English should be chosen. It could be argued that people with experience in acoustics listens for other criteria than other people, but since the term loudness should be understood by everybody there will be made no distinction between people who have experience in acoustics and people who have not.
4.4
Reproduction of Test Files
There are different demands to be fulfilled for running the listening test, these are: • The compared impulse sounds shall be perceived to come from the frontal direction of the test subject, in order to resemble a natural sound event.
46
4.5. PROCEDURE FOR SELECTING TEST FILES.
– This can be obtained by either using a loudspeaker in front of the test subject, or by using binaural sound presented through headphones. • Some of the compared impulse sounds are recorded in a echoic room. – The sounds should therefore be either reproduced by a loudspeaker in an anechoic room or by headphones, to avoid the influence of the room twice • The presented sounds should not impair the hearing of the listener, and the level should therefore be adjusted to a comfortable listening level, which is assumed to be a peak SPL of ≤ 100 dB and a LAeq of ≤ 70 dB. The check of level is described in appendix E.2. Headphones Binaural sound in headphones is a technique usable to give the listener the impression that a sound can come from all three dimensions, the tecnique is therefore also known as 3D-sound. The tecnique filters the signal with Head Related Transfer Functions (HRTF) in order to create the spatial sensation. The HRTF’s are obtained by impulse response recordings on either an artificial head or a person. The problem is that only persons with the exact same head dimentions as the head used for obtaining the HRTF’s will percieve an exact 3D-impression. Loudness judgements applied to test subjects by inaccurate HRTF’s will only give approximate loudness. The advantage of using binaural technique in headphones is that the listening test can be executed in every silent room.
Loudspeaker The problem with using a loudspeaker is that the listener and the loudspeaker need to be located in an anechoic room, since some of the compared impulse sounds are recorded in an echoic room, and if reproduced in a echoic room the listener will get the impression from both rooms. The frequency response of the path between the sound source and to the ear of the listener should also be approximately linear. It is chosen to use the loudspeaker and the anechoic room for the listening test.
4.5
Procedure for Selecting Test Files.
When choosing the test files the difference between the stimuli should be large enough to generate a perceivable loudness difference (if possible) but still small enough in order not to make the task to obvious. This means that choosen stimuli should not result in a 100 % perceived equally stimuli, neither stimuli perceived equally 50 % of the time (meaning that they are guessing). The sound samples used as stimuli for the listening test are all chosen more or less based on the same procedure. A flow chart showing the selection of the sound samples can be seen in figure 4.1. The first step is to normalize the data to the three energy normalizations Leq , LAeq and LCeq . This normalization ensures that all sound samples have the same equivalent continous A-, C- or unweighted sound pressure level and is performed as described in appendix C. This normalization
47
b
CHAPTER 4. LISTENING TEST
# $
# $
! "
! "
! "
# $
Figure 4.1: A flow chart showing the procedure of selecting the sound samples for the listening test.
results in three blocks of data each consisting of files normalized to the same energy, where the only difference between each block of data is an amplitude scaling dependent on each file. From each pool of energy normalized data, sound samples for each of the three test parts should be chosen, meaning data for the test of B-duration, test of peak SPL and test of frequency content. For each of the energy normalized blocks of data the further selection of sound files is identical and can be seen in the flow chart in figure 4.2 For each of the three test parts (B-duration, peak value and frequency content) only the parameter under test should be varied and the two other should be kept constant. This means that for the test of influence of B-duration the peak level and frequency content should be kept constant, etc. For each test part of the different parameters and the different energy normalizations it is chosen to use recorded data from two different types of plates for the test of B-duration and peak value or from two different environments for the test of frequency content. This is done in order to have stimuli presented in the test from two different conditions rather than only one. In this way the same parameter (B-duration, peak level and frequency content) can be evaluated in two situations to thereby give a wider perspective of the influence of the parameter. For the test of the influence of the B-duration and peak level, the two conditions are chosen to be a difference in the type of the plate (metal and wood). For the test of the influence of the frequency content the two conditions are chosen to be a difference in the environment (diffuse field and near field). As stated in section 3.3 a simple way of controlling the frequency content is simply by controlling the setup and the environment. Keeping the frequency content constant for the files means that the sound samples should be chosen from the same environment and the same setup. For a difference in frequency content the sound samples should therefore be chosen from different types of plates but still from the same environment to keep the B-duration and peak level approximately constant. A way to check the value of the parameters B-duration and peak level and keep them either constant or varied is to plot the data in a scatter plot with the B-duration on one of the axis and the peak level on the other axis. Based on such a plot a visual inspection of the data, with the limits described in section 3.6, can serve as a foundation for the selecting of sound samples for the test.
48
4.6. EXECUTION OF THE LISTENING TEST
% & ' ( )* + ,- . * & / 0 ( & .1 - , 1 0 * 1* + 1
2 - 3 4 5 1 5- 3 6
2 - 3 4 51 5 - 3
* & . ; 5* ) 4
% / & 1 1 * . ( )- 1
% / & 1 1* . ( )- 1
? 51 0 ) 5 ' 51 +
? 51 0 ) 5' 51 +
2 0 - - + 53 @ A
2 0 - - + 53 @ A
+ & ' ( )* +
+ & ' ( )* +
% ( * / 1 .- @ .& '
- ,
% ( * / 1 .- @ .& '
- ,
/ 0 - + * 3 + & ' ( )* +
/ 0 - + * 3 + & ' ( )* +
2 0 - - + 53 @ 1 0 * , 5 3 & )
2 0 - - + 5 3 @ 1 0 * , 53 & )
+ & ' ( )* +
+ & ' ( )* +
B & )54 & 1 * 4 1 0 . - : @ 0
B & ) 54 & 1 * 4 1 0 . - : @ 0
& : . & ) 5 3 + ( * / 1 5- 3
& : . & ) 53 + ( * / 1 5- 3
Figure 4.2: Flow chart showing the selection of files for the test parts. These steps are performed after the sound files have been energy normalized.
Further the frequency content of the sound files is checked with spectrograms plotted in the same was as described in section 3.4.6. These two type of plots, spectrograms and scatter plots, serve as the background for choosing the signals for the test. Finally the chosen signals are validated through a aural inspection of the sound stimuli performed by the authors. The selecting of the files for the three test parts including plots of data is described in further details in appendix D.
4.6
Execution of the Listening Test
In this section the execution of the listening test will be described.
Training Session An instruction to the test for the subjects in the listening test is constructed, so the test subjects can fully understand their task without an oral instruction that would differ from subject to subject and thereby bias the subjects. Even the best written instructions can cause some uncertainty about the
49
b
CHAPTER 4. LISTENING TEST
task for the test subjects, which can cause that the first presented samples in the listening test are useless when analyzing the results. To minimize this and to familiarize the test subjects with their task a short training session is constructed which contains 2 pairs of test files from all three parts of the test.
Equipment Setup In order to keep the position of the subject and the head direction constant throughout the listening test the subjects are seated in a chair with an attached device (head rest) to help the subject keep the same head position throughout the test. The Genelec 1031A is chosen as the loudspeaker since it has a flat frequency characteristic (see figure 4.3) and a built in amplifier with a high slew-rate (80 V/µs).
Figure 4.3: Frequency response for the chosen loudspeaker. The upper curve group shows the horizontal directivity characteristics of Genelec 1031A in its vertical configuration measured at 1 m. The lower curve is a 1/3 octave band power response, measured in an IEC approved reverberation chamber. [Genelec, 1997]
To ensure that the frequency response of the reproduction system (from the computer where the stimuli are played and to the position of the test subject) is approximately flat a graphic equalizer using one-third octave bands has been added to the system. The procedure to equalizae the reproduction system chain is described in appendix E.2.
Check of LAeq and Peak SPL In order to ensure that the test subject will not receive hearing impairment from participating in this experiment the LAeq and peak SPL have been checked for all sounds in the listening test. This has been done by setting up a microphone positioned in the place of the middle of the head of the subject with the subject absent. This is described in appendix E.2. In figure 4.4 is shown the LAeq and peak SPL for all test files. It can be seen that all LAeq values are below 70 dB and that the maximum peak SPL is 100.4 dB. This is considered not to induce any hearing damage to the test subjects. 50
4.7. ANALYSIS OF PRESENTED STIMULI
Peak SPL [dB]
100
95
90
85
80
B−duration
Peak Value
Frequency Content
B−duration
Peak Value
Frequency Content
68 66
LAeq [dB]
64 62 60 58 56 54
Figure 4.4: Peak SPL and LAeq of the recorded signals to the listening test.
Graphical User Interface & Sound Control The listening test is executed through MATLAB which can play sound through the loudspeaker connected by the sound card and through a graphical user interface show information to the subject and record user action by a touch screen. The setup of the listening test is further described in appendix E.1 and the setup of the graphical user interface is described in appendix F.
Pilot Test During the design of the listening test several informal pilot tests have been conducted to check the test instruction of the test, the graphical user interface, the duration between the presented signals, the duration of the test parts, the complexity of the tests and naturally the listening test itself.
Completion of Test The tests was carried out over a week and in all 19 test subjects participated.
4.7
Analysis of Presented Stimuli
After the listening test was successfully carried out, the data could be analyzed. It was noticed that the peak SPL in figure 4.4 differed from what was expected. The stimuli presented during the listening test was recorded and the parameters: B-duration, the peak level, the frequency content and LXeq was calculated for this recorded stimuli. In section I.1 a comparison is made between these calculated parameters for both the original signals and for the recorded signals. In the following the terms: “original test files” and “recorded test files” will be used when referring to respectively the signals originally recorded to use as stimuli in the listening test and the signals recorded from the loudspeaker. In the next section the error causing the variation is localized and described. 51
b 4.7.1
CHAPTER 4. LISTENING TEST
Error Localization
To localize the error the output of each part of the reproduction chain is checked. This is done by playing an ideal impulse in MATLAB and record the output from each part of the chain, which is the output of the soundcard, the output of the equalizer and the output of the loudspeaker recorded in a position equal to the listeners position during the listening test. This is described in appendix E.3.
Impulse Response The ideal impulse in MATLAB is a single sample with a value of one or minus one preceded and followed by a series of zeros, this can be seen in the first column of plots in figure 4.5. The output of each stage is displayed in figure 4.5. The top row of plots are the response to the positive impulse and the bottom row of plots are the response to the negative impulse through the reproduction chain.
0.5
0.5
0 −0.5
−1 0.5 Time [ms]
1
0 −0.5
−1 0
−1 0
0.5 Time [ms]
1
0
1
0.5
0.5
0.5
−0.5
Amplitude [V]
1
Amplitude [V]
1
0
0 −0.5
−1 0.5 Time [ms]
1
0.5 Time [ms]
0.5 Time [ms]
1
0
0.5 Time [ms]
1
0
0.5 Time [ms]
1
0.5
0
−1 0
0
−0.5
1
−0.5
−1 0
Genelec output 0.5
Sound Pressure [Pa]
0
Amplitude [V]
0.5
−0.5
Amplitude [V]
After equalizer 1 Sound Pressure [Pa]
After soundcard 1
Amplitude [V]
Amplitude [V]
In MATLAB 1
0
0.5 Time [ms]
1
0
−0.5
Figure 4.5: Impulse response in MATLAB and the output of this through different stages in the reproduction system.
The next column of plots displays the output from the soundcard at the computer played through MATLAB. The third column of plots shows the output from the graphic equalizer, and the last column of plots shows the output from the loudspeaker recorded in a position equal to the position of the listeners heads during the listening test. When comparing the three first columns of plots with the plots in the last column it should be noticed that the y-axis is different. For the first three columns of plots the recording system is directly connected to the output of each stage, which means that the measured output is a voltage value, see figure E.9 in appendix E. For the last column the plots shows what is recorded by a microphone from the loudspeaker. The output from the microphone records a sound pressure value in [Pa]. When comparing the plots it is noticed that there is a phase shift of 180 ◦ between the ideal MATLAB impulse and the output of the soundcard and again between the output of the graphic equalizer and the recording of the loudspeaker. Since a phase shift of 180 ◦ is just a change in polarization it is of no importance. 52
4.7. ANALYSIS OF PRESENTED STIMULI
The change created by the sound card and the graphical equalizer are not crucial since the change is small and mainly due to a filtering process, except for the change in amplitude between the positive and the negative impulse. This is probably due an error in the way the sound card convert a -1 and +1 played in MATLAB. The distortion caused by the amplifier-loudspeaker-room-microphone stage is very damaging to the waveform of impulsive signals, since there is a great amount of overshoot, both before and after the peak. The problem is caused by the mechanical part of the loudspeaker. It is tried to move the membrane of the loudspeaker from its equilibrium position to its maximum position in 1/51200=19.53 µs, which naturally is not possible. A different amplifierloudspeaker system was also measured but a similar response was obtained. The impulse response is also quite fair for reproducing “normal” signals but not sufficient for reproducing impulsive signals, but since the test signals are not ideal impulses the error will not be so significant as with ideal impulses. In figure 4.6 the effect of convolving the recorded impulse response with test stimuli is investigated. The chosen set of stimuli is convolved with the recorded impulse response output of the loudspeaker displayed in the plots in the fourth column in figure 4.5. This should theoretically result in a signal similar to the recorded version of the test stimuli. Convolved with positive impulse
Original
Convolved with negative impulse
Convolved with negative impulse
0.25 B−duration = 193 ms Peak Value = 82.4 dB
0.2
B−duration = 189 ms Peak Value = 79.8 dB
0.15
B−duration = 184 ms Peak Value = 79.9 dB
B−duration = 194 ms Peak Value = 82.4 dB
B−duration = 220 ms Peak Value = 77.2 dB
B−duration = 220 ms Peak Value = 77.2 dB
0.2
0.15
0.15
0.1
0.1
0.05
0.05
0
0
−0.05
−0.05
−0.1
−0.1
0.15 0.1
0.1
0.05
0.05
0.1
0.1 Sound Pressure [Pa]
Sound Pressure [Pa]
Convolved with positive impulse
Original
0.25
0.05 0
0
0
−0.05 −0.05
−0.05
−0.1
−0.1
−0.15
−0.15
0.05 0 −0.05 −0.1
−0.1
−0.15
−0.15 −0.2
−0.2 −0.25
−0.25 0
200 Time [s]
400
0
200 Time [s]
Stimuli A
400
0
200 Time [s]
400
0
200 Time [s]
400
0
200 Time [s]
400
0
200 Time [s]
400
Stimuli B
Figure 4.6: Two original signals convolved with the recorded impulse response of the reproduction system. Notice that for each plot the y-axis is scaled to fit the peak SPL of each signal to ease the comparison of the signals. The first and fourth plot are the original signals chosen from a compared pair of stimuli from the frequency content test part. The first plot shows a recording with the metal plate and the fourth plot is a recording with the wood plate both from the diffuse field recording and energy normalized based on Leq . The impulse response of the reproduction system used for convolving is displayed in the fourth column in figure 4.5.
This set of stimuli used to convolve with the impulse responses is chosen because they are highly affected by the reproduction chain in both peak value and B-duration. The first and the fourth plot in figure 4.6 are the original signals and the following plots are respectively the signals convolved with the recorded positive and negative impulse response of the reproduction system. In each plot is displayed their B-duration and peak value. The plotted original stimuli is a set of stimuli from the frequency content test part, which means that their peak value and B-duration should be approximately similar. This means that their difference should be within 1 dB for the peak value and within 10 ms for the B-duration as described in section 3.6 in chapter 3. This is fulfilled for the original plots, but convolved with the recorded impulse responses the difference in B-duration is between 5 ms and 36 ms and the difference in peak value is between 2.6 dB and 2.7 dB. This is not satisfactory. This could also indicate that the current method for calculating B-duration is not very noise robust.
53
b
CHAPTER 4. LISTENING TEST
Sound Pressure [Pa]
Figure 4.7 displays several subsequent recordings of the same signal played by the loudspeaker and recorded identically. As can be seen the signals are not entirely identical. The value of their peak are for instance not identical, but yields the following values (in the same order as the figure): 85.8, 84.7, 84.9, 85.9, 85.7 and 85.2. This change in peak values for identical recordings of the same played impulse can be due to a too low sampling frequency and background noise. In all plots a second impulse starting at 2 ms can also be observed. This small impulse is probably due to reflections from the touch screen, which suit well with the extra traveled distance since 2 ms is equal to approximately 70 cm. 0.4
0.4
0.4
0.3
0.3
0.3
0.2
0.2
0.2
0.1
0.1
0.1
0
0
0
−0.1
−0.1
−0.1
−0.2
−0.2
−0.2
Sound Pressure [Pa]
0
2
4
0
2
4
0
0.2
0.2
0.2
0
0
0
−0.2
−0.2
−0.2
−0.4
0
−0.4
2 4 Time [ms]
0
−0.4
2 4 Time [ms]
0
2
4
2 4 Time [ms]
Figure 4.7: Time plot of several subsequently outputs of the loudspeaker recorded in the listeners position when playing an ideal impulse in MATLAB.
An impulse response of the amplifier-loudspeaker-room-microphone in the position of the listener is also obtained by an MLS sequence, see appendix E.3. The result of this measurement is displayed in figure 4.8 and if compared to the plots in figure 4.7 it can be seen that they are quite similar which indicates that the reproduction problem is in the amplifier-loudspeaker-room-listener stage.
Impulse Response
0.2 0.1 0 −0.1 −0.2 10
12 14 Time [ms]
16
Figure 4.8: Impulse response of the loudspeaker-room-microphone in listener position, see appendix E.3.
54
4.7. ANALYSIS OF PRESENTED STIMULI
Frequency Response Figure 4.9 shows a FFT of each plot in figure 4.5. The top plots shows FFTs of an ideal impulse which give an ideal flat frequency response as described in section 2.1.1 on page 5. FFT of positive impulse
FFT of negative impulse
0.1
1
After soundcard
100
90
80
0.1
1
10
100
90
80
0.1
1
10
100
90
80
0.1
1 Frequency [kHz]
10
90
80
10
After equalizer
After soundcard After equalizer
In MATLAB
90
80
Genelec output
100
Genelec output
In MATLAB
100
0.1
1
10
0.1
1
10
0.1
1
10
100
90
80 100
90
80 100
90
80
0.1
1 Frequency [kHz]
10
Figure 4.9: FFT of the recording of the output of different stages in the reproduction system when playing respectively a positive and negative impulse in MATLAB in the computer, see figure 4.5.
The next row of plots in figure 4.9 is the frequency response of the sound card. This is approximately flat for frequencies between 40 Hz to 16 kHz. Frequencies outside this range are attenuated. For the high frequencies it is the antialiasing filter in the soundcard which gives this response. The sampling frequency of the test signals is 51.2 kHz, but the sampling frequency of the sound card is 44.1 kHz. For the low frequencies it can be due to the fact that the soundcard can not emit DC. The third row of plots is the frequency response of the system after the graphical equalizer. The setting of the graphical equalizer is displayed in figure 4.10. The procedure of adjusting the graphical equalizer is described in appendix E.2. The majority of the one-third octave bands are in neutral position, the most changes is in the bands with center frequency 80 Hz, 100 Hz and 400 Hz. The change is respectively approximately -1.5 dB, -1 dB and 5 dB.
Figure 4.10: Photo of the setting of the graphical equalizer. The center frequency of the lowest one-third octave band is 25 Hz and the center frequency of the highest band is 20 kHz. The shown setting can amplify -6 dB to 6 dB for each one-third octave band.
When comparing the second and third row in figure 4.9 it can also be seen that the frequency response for the frequency area 80 Hz - 100 Hz is attenuated. The frequency response around 55
b
CHAPTER 4. LISTENING TEST
300 Hz is increased, which is a little odd since the change should be around 400 Hz according to the graphical equalizer. The fourth row of plots in figure 4.9 is the frequency response of the total reproduction chain, from MATLAB to the position of the listener. Compared to the other frequency responses this response contains numerous ripples which originate both from the room response1 and the amplifier-loudspeaker response, which is shown in figure 4.3. Generally the frequency response only deviate ± 3 dB in the frequency range 100 Hz to 16 kHz, which should not be devastating to the results since most of the frequency contents are located in this frequency range.
Sampling Frequency
Sound Pressure [Pa]
In figure 4.11 to figure 4.14 is investigated whether the sampling frequency has been sufficient. Figure 4.11 shows a zoom of the stimuli energy normalized through the LAeq used for the Bduration test part. The zoom shows a very short time window surrounding the peak value. 1
1
1
0.5
0.5
0.5
0.5
0
0
0
0
0
−0.5
−0.5
−0.5
−0.5
−0.5
−0.5
−1
−1
−1
−1
−1
−1
1
1
1
0.5
0.5
0
−1.5
−1.5 −100
0 100 Time [µs]
−1.5 −100
0 100 Time [µs]
−1.5 −100
0 100 Time [µs]
−1.5 −100
0 100 Time [µs]
−1.5 −100
0 100 Time [µs]
−100
0 100 Time [µs]
Figure 4.11: Zoomed time plot around the peak value of the A-weighted normalized stimuli used for the B-duration test part.
Figure 4.12 and figure 4.13 shows similar plots for stimuli energy normalized through the LAeq but used respectively for the peak value test part and for the frequency content test part. For all three figures it seems that the sampling frequency is sufficient since most plots display an approximately smooth curve around the peak. However for some of the plots a higher samplings frequency would be preferred, for example the sixth plot in figure 4.12 has a 0.6 Pa drop after the peak value. In figure 4.14 changes in sampling frequency are analyzed. the left plot in figure 4.14 shows the samples from the original signal plotted as circles which is the first plot in figure 4.13. Interpolation between each sample has been performed followed by addition of white noise, which resulted in the plotted line. This results in a difference of 0.5 dB between the new position and the position of the original peak value. In the right plot in figure 4.14 is illustrated what would be the effect if the sampling is displaced half a sampling period in time. An imaginary sample (*) is placed between the two samples with the highest absolute value. The sample is placed so a smooth curve still can be drawn as illustrated with the black line. The difference in peak value is also here 0.5 dB, which means that the sampling frequency is assumed to be suffiecient. The Leq of the noise floor has been calculated to 30.0 dB, which can have some influence on the parameters. If the noise is a constant value added to all signals the influence will be insignificant, 1 Since
the listening test is executed in the anechoic room there should theoretically be no room response, which means that the room response is mainly due to the equipment in the room.
56
Sound Pressure [Pa]
4.7. ANALYSIS OF PRESENTED STIMULI
3
3
3
3
2.5
2.5
2.5
2.5
3
3
3
3
2.5
2.5
2.5
2.5
2
2
2
2
2
2
2
2
1.5
1.5
1.5
1.5
1.5
1.5
1.5
1.5
1
1
1
1
1
1
1
1
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0.5
0
0
0
0
0
0
0
0
−0.5
−0.5
−0.5
−0.5
−0.5
−0.5
−0.5
−0.5
−1
−1
−1
−1
−1
−1
−1
−1
−1.5
−1.5
−1.5
−1.5
−1.5
−1.5
−1.5
−1.5
−2
−2
−2
−2
−2.5
−2.5
−2.5
−2.5
−100 0 100 Time [µs]
−100 0 100 Time [µs]
−100 0 100 Time [µs]
−100 0 100 Time [µs]
−2
−2
−2
−2
−2.5
−2.5
−2.5
−2.5
−100 0 100 Time [µs]
−100 0 100 Time [µs]
−100 0 100 Time [µs]
−100 0 100 Time [µs]
Sound Pressure [Pa]
Figure 4.12: Zoomed time plot around the peak value of the A-weighted normalized stimuli used for the peak value test part. 1.5
1.5
1.5
1.5
1
1
1
1
1
0.5
0.5
0.5
0.5
0.5
0.5
0
0
0
0
0
0
0
−0.5
−0.5
−0.5
−0.5
−0.5
−0.5
−0.5
−0.5
−1
−1
−1
−1
−1
−1
−1
−1
−1.5
−1.5
−1.5
−1.5
−1.5
−1.5
−1.5
−1.5
1.5
1.5
1.5
1.5
1
1
1
0.5
0.5
0
−100 0 100 Time [µs]
−100 0 100 Time [µs]
−100 0 100 Time [µs]
−100 0 100 Time [µs]
−100 0 100 Time [µs]
−100 0 100 Time [µs]
−100 0 100 Time [µs]
−100 0 100 Time [µs]
Figure 4.13: Zoomed time plot around the peak value of the A-weighted normalized stimuli used for the frequency content test part. 0.5
0.5 Original Org. + noise
Original Changed sample
Sound Pressure [Pa]
0
Sound Pressure [Pa]
0
−0.5
−1 −100
−0.5
−80
−60
−40
−20
0 Time [µs]
20
40
60
80
100
−1 −100
−80
−60
−40
−20
0 Time [µs]
20
40
60
80
100
Figure 4.14: Zoomed time plot (circles) from the first plot in figure 4.13. In the left plot noise has been added and in the right plot an extra sample as been added indicated by a star (*).
since it will only create an offset, but unfortunately the noise is varying around zero. This results in the problem that a positive value can be added to the peak value, causing the limit for calculating the B-duration to be raised. Noise with a negative value could by chance be added to the point defining the lower limit for the B-duration to be lower/higher, which leads to a longer/shorter B-duration. The opposite has naturally the same deteriorating effect.
57
C HAPTER Evaluation of Results from Listening Test
5
This chapter is concerned with the interpretation of results from the listening test. Before analyzing the results from the listening test the results from the catch trials are investigated in order to see if there should be a foundation for excluding some test subjects. All subjects have preferred the stimuli with highest energy for all the catch trials, except one who has chosen different in one of the catch trials. There have therefore been found no evidence for excluding subjects on this basis. To investigate whether there is any significance difference between the results, these are evaluated with statistical methods.
5.1
Statistical Method
During the listening test, subjects have listened to pairs of stimuli where they had to choose the one they perceived louder, or said in another way, the one they preferred in terms of loudness. A common statistical method used for testing for significant difference in preference for a given sample case is the sign test. The listening test consists of n identical independent trials for the different parts of the test for every subject. There are two possible outputs of every trial since these consists of comparing pairs of stimuli, where only one of the stimuli can be preferred. The probability of choosing the preferred one is p =P(preferred) and the probability of choosing the other one is q = 1 − p =P(not preferred). If it is assumed that there is no difference between the presented stimuli it can be assumed a priori that the probability of choosing one or the other stimuli is equal, meaning that p = q = 0.5. The probability of choosing one stimuli over another does not change from trial to trial. Due to these facts the data can be assumed to originate from a binomial probability distribution [Anderson et al., 2003, p. 199]. For a large sample case, meaning a sample size greater than 20 (n ≥ 20), and where the probabilities of choosing one or the other outcome are close to 0.5 (pn and qn are both equal 59
b
CHAPTER 5. EVALUATION OF RESULTS FROM LISTENING TEST
or greater than 10), the binomial distribution can be approximated by a normal distribution, [Gravetter and Wallnau, 1985]. The value of 10 is more a guidance value rather than a strict demand. The main procedure for the statistical test is as follows: 1. State the null hypothesis and the alternative hypothesis. For the null hypothesis it is assumed a priori to the test that there is no significant difference between the two samples under test, but differences just occurs by chance. The alternative hypothesis will then say that there is a significant difference between the two samples. 2. Calculate the test statistics. 3. Find the rejection rule, which means to find the distribution with a certain mean and standard deviation that the data would follow if the null hypothesis is true. From this distribution is found a value that describes the probability of a sample to be located in the extreme area of the distribution. 4. Compare the calculated test statistics to the value found from the distribution and judge whether to accept or reject the null hypothesis. The test hypothesis for the statistical test is: H0 : The probability of choosing one stimuli over another is equal to chance (p = 0.5). This means that the subjects cannot perceive any significant difference between the compared stimuli in terms of loudness. Ha : The probability of choosing one stimuli over another is significant different from chance (p 6= 0.5). This means that the subjects can perceive a significant difference between the compared stimuli in terms of loudness. For this thesis the sign test with normal approximation to the binomial distribution will be used as the statistical method since the sample case of the size n is assumed to be large, and the probability of choosing the preferred stimuli p are equal to the probability of choosing the other stimuli q. The mean value µ and standard deviation σ can therefore be approximated using following equations, [Gravetter and Wallnau, 1985]: mean: µ = pn √ standard deviation: σ = npq
(5.1) (5.2)
For this test following values can be found: p = P(loudest) = 0.5
(5.3)
q=
P(softest)
= 0.5
(5.4)
n=
19 · 4
= 76
(5.5)
= 38
(5.6)
= 4.3589
(5.7)
µ= σ=
60
pn √ pqn
5.2. STATISTICS ON B-DURATION
where p indicates the probability of choosing the louder stimuli, q indicates the probability of choosing the softer stimuli, n are the size of the sample case (subjects · repetitions per subject) and µ and σ is respectively the mean and the standard deviation found in equation (5.1) and (5.2). For the normal approximation to the binomial distribution each sample X has a corresponding z-score. This z-score, also referred to as the test statistics, can be calculated as follows: X − µ X − pn = √ z= (5.8) σ npq Physically the z-score describes how many standard deviations the value X is away from the mean value for a given normal distribution. Therefore, the z-score can be related to the area under the distribution which again can be related to the probability of a given X value will occur. The purpose of the statistical test is to check whether this sample X is likely to occur or if it is located far from the mean value where there is a very little probability of such a sample to occur given that it is from that distribution with a given mean and standard deviation. To be able to compare probabilities with any statistical significance a significance level (α) most be chosen. This significant level is related to the most extreme area under the distribution which is also known as the critical region. The probability of a sample to be located in the critical region can be connected to a z-score for this most extreme area. This z-score, relating to the probability of a sample to be located in this region, can be found from a normal distribution table and is referred to as z-critical. In this way a z-critical can be found for the critical region of the distribution and directly be compared to the calculated z-score. If the z-score is under the distribution but outside the critical region, this indicates that the sample X is close to the mean value and likely to originate from the given distribution. If the z-score is in the critical region it means that this sample X is very unlikely to occur and therefore is assumed to originate from another distribution with another mean and standard deviation. For this statistical test a significance level of α = 0.05 has been chosen. The z-critical for this α has been found in a normal distribution table in [Anderson et al., 2003] to: z-critical = 1.96. The rejection rule for the test is defined as follows: Reject H0 if z > z-critical or if z < −z-critical and accept Ha Otherwise H0 cannot be rejected and must be accepted Whether H0 should be rejected or not for the results of the listening test are investigated in the following sections.
5.2
Statistics on B-duration
For the part of the test concerning the influence of B-duration on the perceived loudness the hypothesis is: Are the subjects capable of hearing a statistical significant difference in loudness for impulse sounds normalized to the same continuous equivalent level and approximately constant peak level and frequency content but with difference in B-duration? 61
b
CHAPTER 5. EVALUATION OF RESULTS FROM LISTENING TEST
H0 : The probability of choosing one stimuli over another is equal to chance (p = 0.5). This means that the subjects cannot perceive any significant difference between the compared stimuli in terms of loudness, when the stimuli only differ in the B-duration. Ha : The probability of choosing one stimuli over another is significant different from chance (p 6= 0.5). This means that the subjects can perceive a significant difference between the compared stimuli in terms of loudness, when the stimuli only differ in the B-duration. The test statistics is calculated for every sample X, which is the number of subjects that preferred a certain stimuli over the other. This are calculated for the two different environments plates for each of the three different energy normalizations. For the case with the metal stimuli and normalized based on LAeq and for the comparison between stimuli 1 and stimuli 2, stimuli 2 were preferred 48 times out of 76 possible. With X = 48 and using the equations (5.8), (5.1) and (5.2) the z-score can be calculated as: z = z =
X − µ X − pn = √ ⇒ σ npq 48 − 0.5 · 76 √ = 2.294 0.5 · 0.5 · 76
(5.9) (5.10)
When this z-score is compared to z-critical as described in the rejection rule above, the following is obtained: 2.294 > 1.96 ⇒ z > z-critical
(5.11)
This means that the null hypothesis must be rejected and the alternative hypothesis accepted, meaning that there is a significant difference in loudness between these two stimuli.
5.2.1
B-duration for the A-Weighted Energy Normalization
The test statistics for the other comparison for the test part regarding B-duration and A-weighted energy normalization can be seen in table 5.1 together with z-critical. Further the last column in the table indicates which hypothesis that must be accepted. setup
comparison
z-score
z-critical
hypothesis
Metal
1 vs 2
2.2942
±1.96
Ha
Metal
1 vs 3
3.4412
±1.96
Ha
Metal
2 vs 3
-3.4412
±1.96
Ha
Wood
1 vs 2
-5.0471
±1.96
Ha
Wood
1 vs 3
-2.7530
±1.96
Ha
Wood
2 vs 3
1.6059
±1.96
H0
Table 5.1: The test statistics and the accepted hypothesis for the data energy normalized based on LAeq .
The statistical result from the table are summarized in figure 5.1 together with a visual indication of how many percentage of the subjects that preferred which stimuli over the other stimuli. The figure shows the results of the comparison between the stimuli pairwise. Each stimuli are energy normalized based on LAeq , both those based on the metal plate (the left part of the figure) 62
5.2. STATISTICS ON B-DURATION
B-duration based on LAeq B−duration [ms] 63
132
stimuli 3
95
stimuli 2
95.9
stimuli 1
95.7
stimuli 1
stimuli 2 Softer
Louder
87.1
peak level [dB]
Louder
51.1
90.4
stimuli 3
95.4
stimuli 2
96
stimuli 1
96.5
stimuli 1
stimuli 3
76.3
stimuli 2 Softer
peak level [dB]
B−duration [ms]
stimuli 3
Figure 5.1: Plots of the results from the test part concerning B-duration with stimuli energy normalized based on LAeq . Left plot shows the stimuli based on the metal plate and to the right stimuli from the wooden plate. The black disks indicate that there is a statistic significant difference between these stimuli in terms of loudness. The value of the B-duration is showed in the x-axis in the top of the figure while the value of the peak level is shown on the y-axis to the right in the figure.
and those based on the wooden plate (the right part of the figure). The B-duration corresponding to each stimulus is showed at the x-axis in the top of the plot. The peak value corresponding to the same stimuli is showed at the y-axis in the right side of the plot. The results of the statistical test are indicated by the colors of the discs. If H0 is rejected and Ha has to be accepted, meaning that there is a significant difference in the perceived loudness, the color of the discs in the comparison is black. If H0 could not be rejected, meaning that there are no significant difference in the perceived loudness, the color of the discs is grey. Tables showing the preferences in percentages can be seen in section J.1 in appendix J. Based on figure 5.1 (and the table J.4) it is possible to rank the stimuli on an ordinal scale ranging from softest to loudest. It can be seen that the subject perceived stimuli 2 as louder compared to stimuli 1, stimuli 3 as louder than stimuli 1 and stimuli 2 as louder than stimuli 3. Based on this it is possible to rank the stimuli from the metal and A-weighted energy normalization part as follows: {stimuli 2} ≥ {stimuli 3} ≥ {stimuli 1}
(5.12)
From figure 5.1 the values of the B-duration for these stimuli can be seen, resulting in the following ranking: {stimuli 2} ≥ {stimuli 3} ≥ {stimuli 1}
{87.1} ≥ {132.0} ≥ {51.1}
(5.13) (5.14)
All perceived with a significant difference. The stimuli from the wood part is perceived as stimuli 1 louder compared to stimuli 2, stimuli 1 louder compared to stimuli 3 and stimuli 3 as louder than stimuli 2. These stimuli can be ranked as follows: {stimuli 1} ≥ {stimuli 3} ≥ {stimuli 2}
(5.15)
63
b
CHAPTER 5. EVALUATION OF RESULTS FROM LISTENING TEST
But without significant difference in perceived loudness between stimuli 2 and stimuli 3. This results in the following rank of value for the stimuli: {stimuli 1} ≥ {stimuli 3} ≥ {stimuli 2}
(5.16)
{63.0} ≥ {90.4} ≥ {76.3}
5.2.2
(5.17)
B-duration for the C-Weighted Energy Normalization
The test statistics for the other parts of the test concerning B-duration is calculated according to the same procedure as explained above. Tables containing the test statistics (like table 5.1) can be seen in appendix J. The results for this part of the test regarding B-duration is shown in figures similar to figure 5.1. The results for the energy normalization based on LCeq is seen in figure 5.2. B-duration based on LCeq B−duration [ms] 113.5
86.7
stimuli 3
96.2
stimuli 2
97.3
stimuli 1
97.2
stimuli 1
stimuli 2 Softer
stimuli 3
Louder
87.1
peak level [dB]
Louder
51
76.3
90.4
stimuli 3
92.7
stimuli 2
93.4
stimuli 1
93.7
stimuli 1
stimuli 2 Softer
peak level [dB]
B−duration [ms]
stimuli 3
Figure 5.2: Plots of the results from the test part concerning B-duration with stimuli energy normalized based on LCeq . Left plot shows the stimuli based on the metal plate and to the right stimuli from the wooden plate. The black disks indicate that there is a statistic significant difference between these stimuli in terms of loudness. The value of the B-duration is showed in the x-axis in the top of the figure while the value of the peak level is shown on the y-axis to the right in the figure.
The stimuli from the part energy normalized based on LCeq and recorded from the metal plate can be ranked with the B-duration values given in figure 5.2 as follows: {stimuli 2} ≥ {stimuli 3} ≥ {stimuli 1}
{87.1} ≥ {113.5} ≥ {51.0}
(5.18) (5.19)
All are perceived with a significant difference. The stimuli from the same energy normalization part but recorded from the wooden plate were perceived as stimuli 2 louder than stimuli 1, stimuli 1 louder than stimuli 3 but with stimuli 2 and stimuli 3 perceived as equal. This makes it impossible to rank these stimuli: {stimuli 2} ≥ {stimuli 1} ⇒ {76.3} ≥ {86.7}
{stimuli 1} ≥ {stimuli 3} ⇒ {86.7} ≥ {90.4}
{stimuli 3} = {stimuli 2} ⇒ {90.4} = {86.7}
64
(5.20) (5.21) (5.22)
5.2. STATISTICS ON B-DURATION
5.2.3
B-duration for the Un-Weighted Energy Normalization
The results for the energy normalization based on Leq is seen in figure 5.3. B-duration based on Leq B−duration [ms] 132
86.7
stimuli 3
95.8
stimuli 2
96.8
stimuli 1
96.6
stimuli 1
stimuli 2 Softer
stimuli 3
Louder
87.1
peak level [dB]
Louder
51
76.3
90.4
stimuli 3
92.7
stimuli 2
93.4
stimuli 1
93.7
stimuli 1
stimuli 2 Softer
peak level [dB]
B−duration [ms]
stimuli 3
Figure 5.3: Plots of the results from the test part concerning B-duration with stimuli energy normalized based on Leq . Left plot shows the stimuli based on the metal plate and to the right stimuli from the wooden plate. The black disks indicate that there is a statistic significant difference between these stimuli in terms of loudness. The value of the B-duration is showed in the x-axis in the top of the figure while the value of the peak level is shown on the y-axis to the right in the figure.
The stimuli from the part energy normalized based on Leq and recorded from the metal plate can be ranked with the B-duration values given in figure 5.3 as follows: {stimuli 2} ≥ {stimuli 3} ≥ {stimuli 1}
{87.1} ≥ {132.0} ≥ {51.0}
(5.23) (5.24)
The stimuli from the same energy normalization part but recorded from the wooden plate with the B-duration values given in figure 5.3 were perceived as stimuli 2 louder than 1, stimuli 3 louder than 2 but stimuli 1 louder than 3, which makes it impossible to rank the stimuli in terms of perceived loudness:
{stimuli 2} ≥ {stimuli 1} ⇒ {76.3} ≥ {86.7}
(5.25)
{stimuli 1} ≥ {stimuli 3} ⇒ {86.7} ≥ {90.4}
(5.27)
{stimuli 3} ≥ {stimuli 2} ⇒ {90.4} ≥ {76.3}
5.2.4
(5.26)
Summarizing the Results of the B-duration Statistics
The statistical tests, concerned with the influence of B-duration, does not show consistency in the loudness perception judged on the basis of this parameter. It can be seen that there are no clear relation between the difference in B-duration and the loudness sensation invoked by the presented stimuli. For the energy normalization based on LAeq and the metal plate there is a significantly perceived difference between all pairs of stimuli. The way they are perceived is with a difference of 87.1 ms 65
b
CHAPTER 5. EVALUATION OF RESULTS FROM LISTENING TEST
between stimuli as the loudest and a difference of 51.5 ms as the softest and with a difference of 132.0 ms as the one perceived between the two other stimuli in terms of loudness. For the results based on the wooden plate for the same energy normalization the results shows that the pair of stimuli with the smallest difference generates the loudest perception and the other stimuli with larger difference in B-duration are perceived softer. This indicates that there is no obvious relationship. This absence of a pattern is repeated for the test results dependent on the B-duration. This could indicate that this parameter might not be a robust parameter to describe the loudness of an impulsive sound. Since the data for the listening test has been modified by the reproduction system as described in section 4.7 it might be difficult to judge these results significantly. Though, for impulses which are normalized to the same energy, in some cases the subjects are capable of perceiving a significant difference in the loudness perception caused by these stimuli, meaning that the equal energy concept judged in terms of loudness does not hold for these presented stimuli.
5.3
Statistics on Peak Level
For the part of the test concerning the influence of the peak level on the perceived loudness the hypothesis is: Are the subjects capable of hearing a statistical significant difference in loudness for impulse sounds normalized to the same continuous equivalent level and approximately constant B-duration and frequency content but with difference in peak level? H0 : The probability of choosing one stimuli over another is equal to chance (p = 0.5). This means that the subjects cannot perceive any significant difference between the compared stimuli in terms of loudness, when the stimuli only differ in the peak value. Ha : The probability of choosing one stimuli over another is significant different from chance (p 6= 0.5). This means that the subjects can perceive a significant difference between the compared stimuli in terms of loudness, when the stimuli only differ in the peak value. The spread in the peak values were originally intended to be larger, but was modified by the reproduction system as described in section 4.7 which caused the difference in peak level between compared stimuli to be less than expected. For this test part stimuli were compared pair wise but not for all possible combinations of the stimuli but only for chosen pairs. This makes it impossible to rank all presented stimuli to each others since they have not all been compared. The test statistics is calculated for every sample X, which is the number of subject that preferred a certain stimuli over the other. This is evaluated for the two different plates and the three different energy normalizations in the same way as for the test for the B-duration influence. The results can be seen in table J.7, J.8 and J.9 in appendix J.
66
5.3. STATISTICS ON PEAK LEVEL
5.3.1
Peak Level for the A-Weighted Energy Normalization
The results are summarized in figure 5.4, 5.5 and 5.6 for the test parts based on respectively the LAeq , LCeq and Leq energy normalization. Peak level based on LAeq B−duration [ms] 69.5
69.1
69.5
B−duration [ms]
69.4
17
17
17
16.9
stimuli 4
98.7
stimuli 3
98.9
stimuli 3
98
stimuli 2
100.2
stimuli 2
98.8
stimuli 1
98.9
stimuli 1
98
stimuli 2
stimuli 3
Louder
Louder
stimuli 1
peak level [dB]
100.1
peak level [dB]
stimuli 4
stimuli 4
stimuli 1
stimuli 2
Softer
stimuli 3
stimuli 4
Softer
Figure 5.4: Plots of the results from the test part concerning peak level with stimuli energy normalized based on LAeq . Left plot shows the stimuli based on the metal plate and to the right stimuli from the wooden plate. The black disks indicate that there is a statistic significant difference between these stimuli in terms of loudness. The value of the B-duration is showed in the x-axis in the top of the figure while the value of the peak level is shown on the y-axis to the right in the figure.
Based on figure 5.4 showing the results from the peak level test part energy normalized based on LAeq and the metal plate it can be seen that stimuli 1 is perceived significantly louder than stimuli 2 and stimuli 3 is perceived louder compared to stimuli 4 but without significant difference. This leads to the following: {stimuli 1} ≥ {stimuli 2} ⇒ {98.9} ≥ {100.2}
{stimuli 3} ≥ {stimuli 4} ⇒ {98.9} ≥ {100.1}
(5.28) (5.29)
Both comparisons have differences in peak level less than 1.5 dB. For the part based on the wooden plate the results is: {stimuli 2} ≥ {stimuli 1} ⇒ {98.8} ≥ {98.0}
{stimuli 4} ≥ {stimuli 3} ⇒ {98.7} ≥ {98.0}
(5.30) (5.31)
All without any significant difference in between the stimuli.
5.3.2
Peak Level for the C-Weighted Energy Normalization
Based on figure 5.5 showing the results from the peak level test part energy normalized based on LCeq and the metal plate it can be seen that stimuli 1 is perceived louder than stimuli 2 and stimuli 4 is perceived louder compared to stimuli 3. Both comparisons are without significant difference. This leads to the following: {stimuli 1} ≥ {stimuli 2} ⇒ {98.9} ≥ {100.3}
{stimuli 4} ≥ {stimuli 3} ⇒ {99.0} ≥ {100.4}
(5.32) (5.33) 67
b
CHAPTER 5. EVALUATION OF RESULTS FROM LISTENING TEST
Peak level based on LCeq B−duration [ms] 69.5
69.1
69.5
B−duration [ms]
69.1
17
16.9
17
17
stimuli 4
98.4
stimuli 3
99
stimuli 3
97.2
stimuli 2
100.3
stimuli 2
98.4
stimuli 1
98.9
stimuli 1
97.2
stimuli 2
stimuli 3
Louder
Louder
stimuli 1
peak level [dB]
100.4
peak level [dB]
stimuli 4
stimuli 4
stimuli 1
stimuli 2
Softer
stimuli 3
stimuli 4
Softer
Figure 5.5: Plots of the results from the test part concerning peak level with stimuli energy normalized based on LCeq . Left plot shows the stimuli based on the metal plate and to the right stimuli from the wooden plate. The black disks indicate that there is a statistic significant difference between these stimuli in terms of loudness. The value of the B-duration is showed in the x-axis in the top of the figure while the value of the peak level is shown on the y-axis to the right in the figure.
Both comparisons have differences in peak level less than 1.5 dB. For the part based on the wooden plate the results are: {stimuli 2} ≥ {stimuli 1} ⇒ {98.4} ≥ {97.2}
{stimuli 4} ≥ {stimuli 3} ⇒ {98.4} ≥ {97.2}
(5.34) (5.35)
The differences between the peak levels is quite small (within 1.5 dB) yet both comparisons are perceived with a significant difference in terms of loudness. The B-duration for these stimuli are also with differences within 1/10 of a mili second, meaning that the B-duration might not have influenced the loudness perception. The fact that there is a perceived difference in loudness could indicate that the loudness perception is caused by another parameter rather than the peak and B-duration. This difference could be in the part of the signal below the -20 dB limit used as the lower limit for the B-duration and in the spectrograms. The signals are based on real recordigs of realistic impact noise and will therefore never be complete identically.
5.3.3
Peak Level for the Un-Weighted Energy Normalization
Based on figure 5.6 showing the results from the peak level test part energy normalized based on Leq and the metal plate the following results can be obtained: {stimuli 1} ≥ {stimuli 2} ⇒ {98.8} ≥ {100.1}
{stimuli 3} ≥ {stimuli 4} ⇒ {98.8} ≥ {100.2}
(5.36) (5.37)
Both comparisons have differences in peak level less than 1.5 dB and are perceived without statistical significance.
68
5.3. STATISTICS ON PEAK LEVEL
Peak level based on Leq B−duration [ms] 69.5
69.4
69.5
B−duration [ms]
69.1
17
16.9
17
16.9
stimuli 4
98.4
stimuli 3
98.8
stimuli 3
97.2
stimuli 2
100.1
stimuli 2
98.1
stimuli 1
98.8
stimuli 1
97.2
stimuli 2
stimuli 3
Louder
Louder
stimuli 1
peak level [dB]
100.2
peak level [dB]
stimuli 4
stimuli 4
stimuli 1
stimuli 2
Softer
stimuli 3
stimuli 4
Softer
Figure 5.6: Plots of the results from the test part concerning peak level with stimuli energy normalized based on Leq . Left plot shows the stimuli based on the metal plate and to the right stimuli from the wooden plate. The black disks indicate that there is a statistic significant difference between these stimuli in terms of loudness. The value of the B-duration is showed in the x-axis in the top of the figure while the value of the peak level is shown on the y-axis to the right in the figure.
For the part based on the wooden plate the results is: {stimuli 2} ≥ {stimuli 1} ⇒ {98.1} ≥ {97.2}
{stimuli 4} ≥ {stimuli 3} ⇒ {98.4} ≥ {97.2}
(5.38) (5.39)
The differences between the peak levels are very small (within 1.5 dB) yet both comparisons are perceived with a significant difference in terms of loudness. The B-duration for these stimuli are also with differences within 1/10 of a mili second, meaning that this parameter might not have influenced the loudness perception. This case is similar to the wood plate recordings based on LCeq .
5.3.4
Summarizing the Results of the Peak Level Statistics
The statistical tests, for analyzing the data of the peak level test, does not show consistency in the loudness judged on the basis of the peak level. It can be seen that there are no clear relation between the difference in peak level and the loudness sensation invoked by these stimuli. The data for the peak level test has, as mention earlier, been modified by the reproduction system, leading to smaller difference between compared stimuli than expected. For the energy normalization based on LAeq and the metal plate there was a perceived significantly difference between the first pair of stimuli, even though the difference between the presented stimuli is less than 0.1 dB and the difference in B-duration is also smaller than 0.5 ms. For the second pair of stimuli there were no significant difference. For the wood plate signals energy normalized based on LCeq there was a statistical perceived difference between both pairs of presented stimuli, while there for the other two pairs of stimuli based on the metal plate and presented with the same energy normalization were no significant difference.
69
b
CHAPTER 5. EVALUATION OF RESULTS FROM LISTENING TEST
For the stimuli presented with the energy normalization based on LAeq and for the wooden plate, the parameters of the stimuli are nearly identical to the same situation with the C-weighted energy normalization, but no significant difference was found between the stimuli normalized based on LAeq while a significant difference were found between the stimuli normalized based on LCeq . The fact that a significant difference is perceived for same pair of stimuli in the peak level test while there are no perceived difference between other pairs of stimuli with approximately the same values of peak level B-duration and frequency content (as plotted in the spectrograms), might indicate that there are confounded variables which have a greater influence on the loudness invoked by the presented stimuli. Though, for impulses which are normalized to the same energy, in some cases the subjects are capable of perceiving a significant difference in the loudness perception caused by these stimuli, meaning that the equal energy concept judged in terms of loudness does not hold for these presented stimuli.
5.4
Statistics on Frequency Content
For the part of the test concerning the influence of the frequency content on the perceived loudness the hypothesis is: Are the subjects capable of hearing a statistical significant difference in loudness for impulse sounds normalized to the same continuous equivalent level and approximately constant B-duration and peak level but with different frequency content? H0 : The probability of choosing one stimuli over another is equal to chance (p = 0.5). This means that the subjects cannot perceive any significant difference between the compared stimuli in terms of loudness, when the stimuli only differ in the frequency content. Ha : The probability of choosing one stimuli over another is significant different from chance (p 6= 0.5). This means that the subjects can perceive a significant difference between the compared stimuli in terms of loudness, when the stimuli only differ in the frequency content. The parameters in the data for the frequency content test part were also modified by the reproduction system as described in section 4.7. This modification caused the spread in B-duration and peak level for some stimuli pair to be greater than intended, which could influence the subjects loudness perception of the stimuli. Though the previous test for the B-duration and peak level did not show consistent results which might indicate that these parameters (B-duration and peak level) only are of little importance when judging the loudness of impulsive sounds. For this test part regarding frequency content, stimuli were compared pair wise but not for all possible combinations of the stimuli but only for chosen pairs, similar to the presentation in the test part regarding peak level. This makes it impossible to rank all presented stimuli to each others since they have not all been compared. The test statistics is calculated for every sample X, which is the number of subjects that preferred a certain stimuli over the other. This is evaluated for the two different recording situations and the three different energy normalizations in the same way as for the test for the B-duration influence. The results can be seen in table J.13, J.14 and J.15 in section J.3 in appendix J.
70
5.4. STATISTICS ON FREQUENCY CONTENT
The results are summarized in figure 5.7, 5.8 and 5.9 for the test parts based on respectively the LAeq , LCeq and Leq energy normalization. All the even number stimuli (stimuli 2 and stimuli 4) are based on the metal plate with the broader frequency range and the odd number stimuli (stimuli 1 and stimuli 3) are based on the wooden plate which mostly contain low frequency.
5.4.1
Frequency Content for the A-Weighted Energy Normalization Frequency content based on LAeq B−duration [ms] 176.8
182
189.2
B−duration [ms] 187.9
62.5
43.6
134.1
117.9
stimuli 4
98.1
stimuli 3
90.5
stimuli 3
96.5
stimuli 2
93
stimuli 2
98.3
stimuli 1
92.9
stimuli 1
99.8
stimuli 2
stimuli 3
stimuli 4
Louder
Louder
stimuli 1
peak level [dB]
90.9
peak level [dB]
stimuli 4
stimuli 1
stimuli 2
Softer
stimuli 3
stimuli 4
Softer
Figure 5.7: Plots of the results from the test part concerning the frequency content with stimuli energy normalized based on LAeq . Left plot shows the stimuli based on the diffuse field recordings and to the right stimuli from the near field recordings. The black disks indicate that there is a statistic significant difference between these stimuli in terms of loudness. The even stimuli (2 and 4) are based on the metal plate and the odd stimuli (1 and 3) are based on the wooden plate. The value of the B-duration is showed in the x-axis in the top of the figure while the value of the peak level is shown on the y-axis to the right in the figure.
Based on figure 5.7 showing the results of the diffuse field recordings energy normalized based on LAeq , it can be seen that stimuli 2 is perceived significantly louder than stimuli 1 and stimuli 4 is perceived louder compared to stimuli 3 all with statistical significance. This leads to the following: {stimuli 2} ≥ {stimuli 1} ⇒ {metal} ≥ {wood} {stimuli 4} ≥ {stimuli 3} ⇒ {metal} ≥ {wood}
(5.40) (5.41)
The difference in peak level is within 1 dB for both comparisons. Further the difference in Bdurations are within 6 ms. In both cases the metal plate with the broader frequency range has been preferred over the low frequency content wooden plate in terms of loudness. For the same part of the test but based on the near field recordings the results are: {stimuli 2} ≥ {stimuli 1} ⇒ {metal} ≥ {wood}
{stimuli 4} = {stimuli 3} ⇒ {metal} = {wood}
(5.42) (5.43)
There was only a perceived significant difference between stimuli 1 and stimuli 2.
5.4.2
Frequency Content for the C-Weighted Energy Normalization
Based on figure 5.8 showing the results of the diffuse field recordings energy normalized based on LCeq , it can be seen that stimuli 2 is perceived significantly louder than stimuli 1 and stimuli 4 is 71
b
CHAPTER 5. EVALUATION OF RESULTS FROM LISTENING TEST
Frequency content based on LCeq B−duration [ms] 195.9
182.3
185.8
B−duration [ms] 182.3
104.8
117.9
134.3
90.4
stimuli 4
92.7
stimuli 3
91
stimuli 3
94.3
stimuli 2
90.9
stimuli 2
96.3
stimuli 1
90.5
stimuli 1
96.1
stimuli 2
stimuli 3
stimuli 4
Louder
Louder
stimuli 1
peak level [dB]
91.1
peak level [dB]
stimuli 4
stimuli 1
stimuli 2
Softer
stimuli 3
stimuli 4
Softer
Figure 5.8: Plots of the results from the test part concerning the frequency content with stimuli energy normalized based on LCeq . Left plot shows the stimuli based on the diffuse field recordings and to the right stimuli from the near field recordings. The black disks indicate that there is a statistic significant difference between these stimuli in terms of loudness. The even stimuli (2 and 4) are based on the metal plate and the odd stimuli (1 and 3) are based on the wooden plate. The value of the B-duration is showed in the x-axis in the top of the figure while the value of the peak level is shown on the y-axis to the right in the figure.
perceived significantly louder compared to stimuli 3 all with statistical significance. This leads to the following: {stimuli 2} ≥ {stimuli 1} ⇒ {metal} ≥ {wood}
{stimuli 4} ≥ {stimuli 3} ⇒ {metal} ≥ {wood}
(5.44) (5.45)
The difference in peak level is within 1 dB for both comparisons. But for the difference in Bdurations there are a difference of 13.6 ms in favor of stimuli 1 (the wooden plate). For both comparisons the metal plate with the broader frequency range has been preferred over the low frequency content wooden plate in terms of loudness with significant difference. For the same part of the test but based on the near field recordings the results are: {stimuli 2} ≥ {stimuli 1} ⇒ {metal} ≥ {wood}
{stimuli 4} ≥ {stimuli 3} ⇒ {metal} ≥ {wood}
(5.46) (5.47)
All the stimuli were perceived with statistical significant difference between them and peak values within 1.6 dB. For stimuli 3 and stimuli 4 there is a difference in B-duration of 43.9 ms in favor of stimuli 4, which might have influence the results of this comparison.
5.4.3
Frequency Content for the Un-Weighted Energy Normalization
Figure 5.9 shows the results from the test part concerning the frequency content and energy normalized based on Leq and the diffuse field recordings. Based on this it can again be seen that stimuli 2 is perceived significantly louder than stimuli 1 and stimuli 4 is perceived significantly louder compared to stimuli 3 all with statistical significance. This leads to the following: {stimuli 2} ≥ {stimuli 1} ⇒ {metal} ≥ {wood}
{stimuli 4} ≥ {stimuli 3} ⇒ {metal} ≥ {wood}
72
(5.48) (5.49)
5.4. STATISTICS ON FREQUENCY CONTENT
Frequency content based on Leq B−duration [ms] 185.8
181.9
B−duration [ms] 184
220.1
134.1
90.4
104.8
117.9
stimuli 4
96.2
stimuli 3
92.2
stimuli 3
96
stimuli 2
91.6
stimuli 2
92.7
stimuli 1
90.8
stimuli 1
94.2
stimuli 2
stimuli 3
stimuli 4
Louder
Louder
stimuli 1
peak level [dB]
89
peak level [dB]
stimuli 4
stimuli 1
stimuli 2
Softer
stimuli 3
stimuli 4
Softer
Figure 5.9: Plots of the results from the test part concerning the frequency content with stimuli energy normalized based on Leq . Left plot shows the stimuli based on the diffuse field recordings and to the right stimuli from the near field recordings. The black disks indicate that there is a statistic significant difference between these stimuli in terms of loudness. The even stimuli (2 and 4) are based on the metal plate and the ode stimuli (1 and 3) are based on the wooden plate. The value of the B-duration is showed in the x-axis in the top of the figure while the value of the peak level is shown on the y-axis to the right in the figure.
For the comparison between stimuli 3 and 4, the difference in peak level is 3,2 dB in favor of stimuli 3 (the wooden plate). Further the difference in B-durations are within 36.1 ms for the same pair of stimuli but in favor of stimuli 4 (the metal plate). Even though there is a difference of more than 3 dB between the peak values of the stimuli, the metal based stimuli with the lowest peak is still preferred over the other stimuli in terms of loudness. For the other comparison the metal plate with the broader frequency range is also preferred over the low frequency content wooden plate. For the same part of the test but based on the near field recordings the results is: {stimuli 2} ≥ {stimuli 1} ⇒ {metal} ≥ {wood}
{stimuli 4} = {stimuli 3} ⇒ {metal} = {wood}
(5.50) (5.51)
There was only a perceived significat difference between stimuli 1 and stimuli 2. In this comparison there are a different in B-duration of 43.7 ms in favor of the metal plate, but a difference in peak level of 1.5 dB in favor of the wooden plate. For the comparison between stimuli 3 and 4 there is no difference.
5.4.4
Summarizing the Results of the Frequency Content Statistics
For the part of the listening test concerning the perceived loudness influenced by changes in the frequency content, there was a perceived significant difference between nearly all presented stimuli. Except for two pair of stimuli, all the signals based on the metal plate were perceived as significantly louder than the signals based on the wooden plate. Based on the results of the statistics concerning this test part, it seems that there is a clear tendency towards the subjects choosing the metal plate with the broader frequency range, which also contains higher frequencies, as the louder compared to the wooden plate which mainly contain low frequencies. 73
b
CHAPTER 5. EVALUATION OF RESULTS FROM LISTENING TEST
The stimuli presented in this part of the test are among those mostly affected by the reproduction system which results in some rather large deviations from the expected values for those parameters that were supposed to be kept constant. From the previous sections 5.2 and 5.3, concerning the influence of the B-duration and peak level, there was found no clear evidence of a relation between the value of the presented parameters and the evoked loudness perception. This might indicate that the influence of these other parameters (B-duration and peak level) does not have a robust and significant influence on the loudness perception of impact signals. Therefore the results from the test part concerning the frequency content can still be valid. The conclusion from the test part concerning the influence of the frequency content is that the stimuli based on the metal plate generates a higher loudness perception compared to the stimuli based on the wood plate. This means that the stimuli with the broadest frequency content were perceived as the loudest.
5.5
Results Across Energy Normalizations
For the listening test three different energy normalizations were used to give the presented stimuli equal acoustic energy. These normalizations, see appendix C, are based on the equivalent continuous sound pressure level calculated for the presented signals with the frequency weightings: A-weighted, C-weighted and un-weighted. The normalizations were performed in order to investigate if the A-weighting function, which is utilized in nearly all standards regarding the assessment of noise, performs better than the C-weighting function, which mainly are used for measures of peak levels and the un-weighted signals, which are more descriptive for the physical properties of the signals. For comparing the different energy normalizations, the number of times a comparison have been found to have a statistical significant difference for all comparisons across the test parts with different parameters, are summarized in table 5.2. Further the table shows the number of times where no significant difference have been found. Energy normalizations
Significant difference
Insignificant difference
Total
LAeq
LCeq
Leq
B-duration
5
4
3
Peak level
1
2
2
Frequency
3
4
3
Total
9
10
8
B-duration
1
2
3
Peak level
3
2
2
Frequency
1
0
1
Total
5
4
6
14
14
14
Table 5.2: The number of statistical significant differences for each energy normalization.
From this table it can be seen that for the energy normalization based on LAeq a significant difference between compared stimuli is found 9 out of 14 times, for the energy normalization based 74
5.6. SUMMARIZING AND DISCUSSION OF THE RESULTS
on LCeq a significant difference between compared stimuli is found 10 out of 14 times and for the energy normalization based on Leq a significant difference between compared stimuli is found 8 out of 14 times. This indicates that most significant differences between compared stimuli has been found with the LCeq energy normalization, second most significant differences between compared stimuli has been found with the LAeq energy normalization and least significant differences between compared stimuli has been found with the Leq energy normalization. However the differences in number of perceived significant comparisons are very small. Further the stimuli presented for each energy normalization is not the exact same signals which also makes a comparison more uncertain. When the stimuli were normalized based on the Leq , the majority of the test subjects had to guess between the presented stimuli (or at least the subjects could not agree on which one was the louder). This could be a indication of that the un-weighted energy normalization are the best suited normalization to describe the impulsive sounds tested in this thesis.
5.6
Summarizing and Discussion of the Results
From the statistical tests performed in the previous sections, it seems that neither the B-duration nor the peak value is a robust parameter for the description of loudness of impulsive sounds. This could indicate that the frequency content is a parameter which has a stronger influence on the loudness perception for impulsive signals compared to the two other parameters, even though the data in the frequency test are affected by the change in peak level and B-duration caused by the reproduction system A signal with a wide frequency range is perceived louder than a signal with more narrow frequency range is however not a new results but, as mentioned in section 2.2.1, supports the common theory as mentioned in [Moore, 2004]. The B-duration is quite easily affected by background noise or electrical noise in the equipment, (at least the way it has been calculated in this thesis). As mentioned in section 4.7 noise could increase spikes or fluctuations in the time signal and thereby causing samples that due to noise excees the -20 dB limit, by which the limits for the B-duration is defined, see section 2.1.3. B-duration will thereby be increased artificially. The peak value, even though it has not been evaluated as thoroughly as first intended, is easily defined, but it raises some demands for the measuring system if a precise measure of the peak value is desired due to the very narrow peaks in impulsive sounds. A parameter which is capable of describing impulsive signals while taking time aspect together with peak relations into consideration might be a suited parameter, but it still needs a strong definition. The B-duration is a parameter that is concerned with both time and level aspects but it is weakly defined in the standard [DS/ISO 10843, 1998]. One issue that could not be emphasized enough, based on the learning process of this thesis, is when dealing with demanding signals like the impulsive signals it is important to use the appropriate equipment and know its limitations. This is important since no measuring equipment is ideal and the nature of impulsive noise requires extra attention.
75
C HAPTER Conclusion
6
The aim of this thesis is, as stated in section 1.1, to evaluate different parameters descriptive of impulsive noise in terms of perceived loudness in relation to the equivalent continuous frequency weighted sound pressure level. The parameters are chosen based on an analysis of measured types of impulsive noise. The perceived loudness sensation evoked by changes in the found parameters is evaluated based on a subjective listening test. Two frequency weighting functions (A- and C-weightings) are used to calculate the frequency weighted equvalent continous sound pressure level together with a unweighted equvalent continous sound pressure level (LXeq ) to evaluate compaired stimuli in terms of equal energy. In the previous chapters the theory behind impulsive sounds, with emphasize on impact noise, has been summarized and recordings of impact noise generated by impacts of combinations of wood and metal has been performed and later analyzed in the perspective of the theory. The impact noise was recorded both in an anechoic room and in an echoic room (at the department of acoustics at AAU). Two recording setups were used in the echoic room, one with the microphone close to the noise source in order to record the main influence of the direct sound and, one with a microphone far away from the noise source in order to record the noise source together with the influence caused by the environment. Based on the analysis the following parameters were chosen for evaluation in the listening test: • B-duration • Peak level • Frequency content Further the compared stimuli presented in the listening test are normalized to have equal acoustic energy, calculated as LAeq , LCeq and Leq . This means that the compaired stimuli are presented with either the same Leq , LAeq or LCeq . The performed listening test utilizes the method of constant stimuli suited for discrimination test between two successively presented stimuli. For obtaining the results from the test subjects the 2IFC method was used for collecting the preferences in terms of loudness as judged by the subjects. 77
b
CHAPTER 6. CONCLUSION
In total 19 subjects participated in the listening test. After the execution of the test it was noticed that the reproduction system was not as ideal as expected which caused a change in the parameters calculated for the stimuli presented to the test subjects. The change was larger for some stimuli than for other. This resulted in that when the parameters should be constant for some sets of the presented stimuli they were not as constant, meaning that they can influence the loudness sensation that should be judged based on varied stimuli. The change in parameters was caused by the reproduction system, mainly the soundcard and the loudspeaker, which modified the stimuli in an unintended and, prior to the test, unexpected manner. If instead the parameters were calculated for signals convolved with the impulse response of the reproduction system the parameter values presented to the test subject are obtained and on this basis stimuli for the listening test can be selected. For the part of the listening test concerning the perceived loudness influenced by changes in the B-duration parameter, there was a significant perceived difference between some stimuli but not for others. There was no clear relationship between the physical value of the B-duration and which comparisons were found to be significant. This indicates that the B-duration (as calculated in this project) is not a robust parameter for describing the loudness sensation evoked by impact sounds (for those presented in this thesis). The way the B-duration is calculated should be based on a smooth envelope of the signal as described in the [DS/ISO 10843, 1998], though this is weakly defined. Based on discussions in section 5.6 it can be concluded that a stronger definition of the envelope is needed and thereby an improved method for calculating the B-duration which might show different results. For the part of the listening test concerning the perceived loudness influenced by changes in the peak value parameter, there was a significant perceived difference between very few stimuli while most of the comparisons showed insignificant differences. This part of the test was highly affected by the reproduction system, resulting in smaller differences in peak value. This compressed range of peak value from 97.2 dB to 100.4 dB for the presented stimuli was therefore also expected to cause many insignificant answers. Since the range of the stimuli was smaller than excepted and the subjects were answering inconsistently it made it difficult (impossible) to range the compared stimuli for the three energy normalizations. This indicates that the range of stimuli presented in this thesis is inadequate to judge the importance of the peak value in terms of perceived loudness. The peak level has howerver a clear relation to the nature of impulsive signals. For the part of the listening test concerning the perceived loudness influenced by changes in the frequency content, there was a significant perceived difference between nearly all presented stimuli, with the exception of two pairs of stimuli. This part of the test was the one most affected by the reproduction system, resulting in some rather large deviations from the expected values for those parameters that were supposed to be kept constant. Even though these deviations were present the influence of the parameters (B-duration and peak) did not seem to have a robust and significant influence on the loudness perception. The conclusion from the test part concerning the influence of the frequency content, as stated in section 5.6, is that the stimuli based on the metal plate generates a higher loudness perception compared to the stimuli based on the wooden plate. This means that the stimuli with the broadest frequency content were perceived as being the loudest. This is however not a new result, but confirms the commonly accepted theories mentioned in section 2.2.1 and by [Moore, 2004], that a wider frequency contents leads to a higher loudness perception. Regarding the different energy normalizations as described in section 5.5, the results showed little difference. Out of the 14 different comparisons within each energy normalization 9 were perceived
78
6.1. FUTURE WORK
significantly different for the LAeq normalization, 10 for the LCeq normalization and 8 for the Leq normalization. This could indicate that the Leq is best suted for the assessment of impulsive noise since most of the subjects could not tell the difference between stimuli presented with equal acoustic energy calculated as the equivalent continous un-weighted sound presure level. Subjects were able to perceive a significant difference in terms of loudness between stimuli presented with equal acoustic energy. This means that the loudness sensation evoked by impulsive signals is not solely judged by their acoustic energy.
6.1
Future Work
Due to the fact that the parameters of the presented stimuli were affected by the reproduction system further tests with a controlled influence of the reproduction system should be performed to confirm the results and to support the conclusions of this thesis. This is mainly concerned with the parameters of B-duration and peak level. The B-duration parameter might be a useful parameter for the description of impulse sounds since it is related both to the time and level aspects of a impulsive sound.
79
Bibliography [Anderson et al., 2003] David R. Anderson, Dennis J. Sweeney and Thomas A. Williams. Modern Buisness Statistics - With Microsoft Excel. South-Western, Thomason Learning, 2003, 1st . edition. ISBN: 0-324-12174-1. [Arbejdsmiljø, 2002] Sebastian Swiatecki. Mangelfuld standard for støj. Arbejdsmiljø, No. 8, August 2002. [Arbejdstilsynet, 2006] Arbejdstilsynet. Beskyttelse mod udsættelse for støj i forbindelse med arbejdet, 2006. URL: http://www.at.dk/sw19871.asp. Downloaded the 26/5-2006. [Beranek and Ver, 1992] Leo L. Beranek and Istan L. Ver. Noise and Vibration Control Enegineering. John Wiley & Sons, Inc., 1992. ISBN: 0-471-61751-2. [Berglund et al., 1999] Birgitta Berglund, Thomas Lindvall and Dietrich H. Schwela. Guidelines for community Noise. World Health Organization (WHO), 1999. [Carter, 1965] N. L. Carter. Effects of Repetition Rate on the Loudness of Triangular Transients. The Journal of the Acoustical Society of America, Vol. 37, No. 2, February 1965, page 308-312. [Carter, 1972] N. L. Carter. Effects of Rise Time and Repetition Rate on the Loudness of Acoustic Transients. Journal of Sound and Vibration, Vol. 21, No. 2, 1972, page 227-239. [Coles et al., 1967] R. Ross, A. Coles, Georges R. Garinther, David C. Hodge and Christopher G. Rice. Hazardous Exposure to Impulse Noise. The Journal of the Acoustical Society of America, Vol. 43, No. 2, 1968, page 336-343. [Directive 2002/49/EC, 2002] The European Parliament and the Council of the European Union. Directive 2002/49/EC of the European Parliament and of the Council of 25 June 2002 relating to the assessment and management of invironmental noise, 2002. http://europa.eu.int/eur-lex/pri/en/oj/dat/2002/l_189/l_ URL: 18920020718en00%120025.pdf. Downloaded the 27/5-2006. [Directive 2003/10/EC, 2003] The European Parliament and the Council of the European Union. Directive 2003/10/EC of the European Parliament and of the Council of 6 February 2003 on the minimum health and safty requirements regarding the exposure of workers to the 80
BIBLIOGRAPHY
risks arising from physical agents (noise), 2003. http://europa.eu.int/eur-lex/pri/en/oj/dat/2003/l_042/l_ URL: 04220030215en00%380044.pdf. Downloaded the 27/5-2006. [DS/IEC 61672, 2003] International Electrotechnical Commission. Electroacoustics - Sound Level Meters - Part 1: Specifications. Dansk Standard, 2003, 1st . edition. [DS/ISO 10843, 1998] International Organization for Standardization. Acoustics - Methods for the description and physical measurement of single impulses or series of impulses. Dansk Standard, 1998, 1st . edition. [DS/ISO 140-6, 1998] International Organization for Standardization. Acoustics - Measurement of sound insulation in buildings and of building elements - Part 6: Laboratory measurements of impact sound insulation of floors. Dansk Standard, 1986, 1st . edition. [DS/ISO 226, 2004] International Organization for Standardization. Acoustics - Normal equalloudness-level contours. Dansk Standard, 2004, 2nd . edition. [DS/ISO 532, 1975] International Organization for Standardization. Acoustics - Method for calculating loudness level. Dansk Standard, 1975, 1st . edition. [Erdreich, 1986] John Erdreich. A distribution based definition of impulse noise. The Journal of the Acoustical Society of America, Vol. 79, No. 4, 1986, page 990-998. [Genelec, 1997] Genelec. Genelec 1031A Bi-amplified Monitoring System - Data sheet. Genelec, 1997. [Glasberg and Moore, 2002] Brian R. Glasberg and Brian C. J. Moore. A Model of Loudness Applied to Time-Varying Sounds. Journal of Audio Engineering Society, Vol. 50, No. 5, May 2002, page 331-341. [Gravetter and Wallnau, 1985] Frederick J. Gravetter and Larry B. Wallnau. Statistics for the Behavioral Sciences. West Publishing Company, 1985. ISBN: 0-314-85241-7. [Hamernik & Hsueh, 1991] Roger P. Hamernik and Keng D. Hsueh. Impulse noise: Some definitions, physical acoustics and other considerations. The Journal of the Acoustical Society of America, Vol. 90, No. 1, July 1991, pp. 189-196. [Henderson & Hamernik, 1986] D. Henderson and R. P. Hamernik. Impulse noise: Critical review. The Journal of the Acoustical Society of America, Vol. 80, No. 2, August 1986, pp. 569-584. [Henderson et al., 1990] Donald Henderson, Malini Subramaniam, Michael Anne Gratton and Samuel S. Saunders. Impact noise: The Importance of Level, Duration, and Repetition Rate. The Journal of the Acoustical Society of America, Vol. 89, No. 3, March 1991, pp. 1350-1357. [ISO 1996-1, 2003] International Organization for Standardization. Acoustics - Description, measurement and assessment of environmental noise - Part 1: Basic quantities and assessment procedures. International Organization for Standardization, 2003, 2nd . edition.
81
b
BIBLIOGRAPHY
[Kryter, 1985] Karl D. Kryter. The Effects of Noise on Man. Academic Press, Inc., 1985, 2nd . edition. ISBN: 0-12-427460-9. [Moore, 2004] Brian C. J. Moore. An Introduction to the Psychology of Hearing. Elsevier Academic Press, 2004, 5th . edition. ISBN: 0-12-505628-1. [Moore et al., 1997] Brian C. J. Moore, Brian R. Glasberg and Thomas Baer. A Model for the Prediction of Thresholds, Loudness, and Partial Loudness. Journal of Audio Engineering Society, Vol. 45, No. 4, April 1997, page 224-239. [NIOSH No. 98-126, 1998] U.S. Department of Health and Human Services (NIOSH). Critera for a Recommended Standard. Occupational Noise Exposure Revised Criteria 1998, 1998. URL: http://www.cdc.gov/niosh/98-126.html. Downloaded the 28/5-2006. [Oppenheim & Schafer, 1999] Alan V. Oppenheim and Ronald W. Schafer. Discrete-Time Signal Processing. Prentice Hall, 1999, 2nd . edition. ISBN: 0-13-754920-2. [Reichardt, 1970] W. Reichardt. Subjective and Objective Meaurement of the Loudness Level of Single and Repeated Impulses. The Journal of the Acoustical Society of America, Vol. 47, No. 6, 1970, page 1557-1562. [Roberto et al., 1984] M. Roberto, R. P. Hamernik, R. J. Salvi, D. Henderson and R. Milone. Impact Noise ant the Equal Energy Hypothesis. The Journal of the Acoustical Society of America, Vol. 77, No. 4, 1985, page 1514-1520. [Snodgrass et al., 1985] Joan Snodgrass,Gail Levy-Berger and Martin Haydon. Human Experimental Psychology. Oxford University Press, 1985. ISBN: 0195035747.
82
