A classification and comparison of vocalizations of captive killer whales (Orcinus orca) MarilynE. Dahlheim NationalMarine Mammal Laboratory,NationalMarine Fisheries Service,NOAA, 7600SandPoint Way N.E., Seattle,Washington 98115

Frank Awbrey Departmentof Biology•San DiegoState University, San Diego,California92182

(Received9 September1981;acceptedfor publication20 April 1982) An acousticalstudywasconductedon captivekiller whales(Orcinusorca)to determineif individualsor groupsdifferedin their vocalizations. Twenty-onedifferentunderwater

vocalization categories produced by 13animalswererecorded withina 40 Hz to 20 kHz bandwidth.Eight acousticalvariables(startingand endingfrequencies of the signal,duration, harmonicinterval,startingand endingfrequencies of the fundamentals, and low and high concentration of energy)weremeasuredfor eachvocalization.MULTIVARIATE DISCRIMINANT analysiswasusedto determinewhetherindividual,oceanarium,or sexualdifferences were detectablein their vocalizations. Resultsshoweddistinctacousticalgroupings by individuals,by oceanariums,and by sex. PACS numbers: 43.80. Lb, 43.80.Nd

INTRODUCTION

Grieg (1906)contributedone of the earliestpublished reportson killer whalesounds,notingtheflutelikecallsfrom younganimalsandroarsfrom oldbullsduringwhalingoperations.Valdez (1961)recorded,but did not extensivelyanalyze, ultrasonicsoundsmadeby theseanimalsin the eastern Atlantic Ocean.Schevilland Watkins(1966)analyzedcalls froma youngcaptivemalein BritishColumbia,notingclicks believed to be used in echolocation and screams assumed to

be for communication.

Much other work has since been con-

Priorinvestigations by theseniorauthorindicatedthat soundproduction in captivitybyOrcinuswasdiurnallyvariableso recordingswere not madeat fixedtimes.With the exceptionof the Sea World recordings,all underwater soundswere recordedwith a CelescoLC-10 hydrophone

anda Nakamichi550cassette taperecorder. • At SeaWorld, A Wilcoxonhydrophoneand a Uher 4400 reel-to-reeltape

recorder'wasused.Thefrequency response ofbothsystems (40 Hz to 19 kHz) was limited by the tape recorders.Our previousrecordingsand a review of the literature showed

killer whales, Orcinus orca, maintained at five west coast

peakenergyin the signalsof O. orcato be below20 kHz, indicatingthat the abovesystemswereadequatefor the proposedresearch.Bothrecordershadtwochannels.Data were recordedon onechanneland simultaneous commentaryrecordedon the other.The identityof the particularanimal makingthesoundwasdetermined bynotingbtibble emission from the blowholeor by the locationof the whalerelativeto thehydrophone.Also, dataonsex,oceanariumlocation,and geographical areaof capturewerenotedfor eachwhale. Onomatopoeic soundtypeswereestablished andexampiesof eachsoundtype were selectedfor eachanimal. A "waterfall"spectrogram of eachworkingtape was made with a SpectralDynamicsmodel 301 real-timespectrum

oceanariums: SeaWorld, SanDiego,CA, fiveanimals;Mar-

analyzer, • sampling 40 msof sound between 0 and10kHz

ineland, Palos Verdes, CA, three animals; Marineworld, ium, Vancouver, British Columbia, Canada, two animals; and Sealand, Victoria, British Columbia, Canada, one ani-

sequentiallyevery50 ms. Pulsesoccurringfasterthan the time per line were resolvedfrom the harmonicinterval of pulsesideband.Due to the samplingrate of the analysis equipment,pulse-repetition ratesof signals,withoutobvious

mal.

harmonic structures, could not be resolved. Effective filter

ductedon thisspecies' sounds(HindsmannetaL, 1966;Singletonand Poulter, 1967;Steineretal., 1979}. The presentacousticalstudywasconductedon captive killer whales,Orcinusorca L., 1758,to testthe hypothesis that individualwhalescan be identifiedby their sounds.If animals at different oceanariums

and of similar

sex are

acousticallyseparable,wild Orcinusindividualsand pods may alsobeacoustically distinguishable. I. MATERIALS

AND METHODS

Underwaterrecordings werecollectedfrom 13 captive

RedwoodCity, CA, two animals;VancouverPublicAquar-

FIG. 1. SampleSl;•ctrogramof

an upscream(50 ms perline).

661

J.Acoust. Soc.Am.72(3),Sept.1982

0001-4966/82/090661-10500-80 ¸ 1982Acoustical Society ofAmerica

601

FIG. 2. Samplespectrogram of a downscream (50 msper line).

bandwidth was 19.5 kHz.

Whennecessary, sonograms weremadewitha Kay Elemetrics Corporation, model7029A,soundspectrograph. • Eight acousticalvariablesfor each soundwere measureddirectly from the spectrograms: minimum frequency (lowestfrequencyobserved); maximumfrequency(highest frequencyobserved);duration (time periodof the signal);

startingfrequencyof thefundamental; endingfrequencyof

thefundamental; andthe frequency intervalbetween harmonics. For soundswith obvious harmonic structure, the

beginningand endingfrequencyof the stressedharmonic was included. Alternatively, if a vocalization was more broadband,with lessobviousharmonicstructure,thebeginningandendingfrequencyof the "stressed" areain the signal (stressed areasappeardarkeron thespectrogram display) was included.

D• = diaz• + daz2+ ... + dipzp, whereDi wasthescoreof thediscriminantfunctioni, thed's wereweightingcoefficients, and the Z's werethe standardizedvaluesof thep discriminating variablesusedin theanalysis.Details of the mathematicalderivationof this procedure can be found in Cooley and Lohnes (1971) and Tatsuoka(1971).In the directmethod,the eightacoustical variableswere enteredinto the analysisconcurrently.The discriminantfunctionswerecreateddirectlyfrom the entire setof variables.In the stepwisemethod,the variableswere selectedfor entry in order of their discriminatingpowerusingthe Rao'sV methodfor theselectionof thevariablesto be includedin thediscriminant(Nie etal., 1975).A variablewas includedonly if its partial multivariateF ratio was larger than a specifiedvalue{ = 1.0).

For statisticalanalysisthe STATISTICALPACKAGEFOR

THESOCIAL SCIENCES (SPSS)wasusedandsubprogram AGGREGATEcalculatedmeans,standarddeviations,and maxi-

II. RESULTS

mumandminimumvaluesfor eachsoundtypefor eachanimal {Nieetal., 1975).All repetitions of eachsoundtypefrom eachanimalweregroupedtogetheranda grandmean,stan-

A. Sound types

dard deviation, and maximum and minimum values were calculated for each sound variable.

The MULTIVARIATE DISCRIMINANT analysis program testedfor interindividual,intergroup,andsexualdifferences in vocalizations. Thesediscriminatingfunctionshad the form

We classified thedifferentsoundsproducedby thecaptive killer whalesinto 21 differenttypes.Somewhalesproducedall 21soundtypes;however,11appeared withgreater frequencythan the others.All the soundswerereadilyassignedby ear to a particulartype and could be identified consistently by otherpeopleeventhoughspectrograms revealedconsiderable variationwithin eachsoundtype. The 11 major soundtypesare listed and describedbelow. The

FIG. 3. Samplespectrogram of a creak(50 ms per line).

662

J.Acoust. Soc.Am.,Vol.72,No.3, September 1982

M:E.Dahlheim andF.Awbrey: Vocalization ofcaptive whales

662

FIG. 4. Samplespectrogram of a whine(50 msperline).

0

1

2

3

4

5

6

7

8

9

10

Frequency (kHz)

FIG. 5. Samplespectrogramof a whistle(50 msper line).

0

'1

2

3

4

5

6

7

8

9

10

Frequency (kHz)

663

J. A½oust.Sec. Am., Vol. 72, No. 3, September 1982

M.E. Dahlheimand F. Awbrey:Vocalizationof captivewhales

663

FIG. 6. Samplespectrogramof a tone(50 ms per line).

valuesreportedfor eachsoundtyperepresentaveragesfrom

Hz) in frequency withtimewasmeasured (Fig.4). The vari-

the records of all animals.

ablefrequencymodulationwithin the whinewasnot includ-

1. Upscream

Upscreams (Fig. 1) wereupwardsweepsin frequency with time. The averagecall duration was 973 ms. These soundstypically began with a rapid burst of broadband pulses.This pulseratethendecreased anda harmonicstructure emerged.Mean startingand endingfrequenciesof the

ed in the analysis.The averagestartingfundamental frequencyof 1435 Hz was slightlyhigherthan that of the upscream, but the endingfrequencyof thefundamentalwas only 1470Hz. Durationaveraged1.2s. The averagepulserepetitionratewas 1200per second.Maximumenergywas concentrated between 3450 and 4417 Hz. Three-hundred whines were recorded from six animals.

fundamental were1220and1865Hz, respectively. Thepulse repetition rate, derived from the harmonic interval, averaged 1050per second.Upscreamsalwayscontainedseveral

strongharmonics. The secondor third harmonicwasusually stressed.Upscreamswere recordedfrom 11 animals, with 1100 usable records.

2. Downscream

Downscreams(Fig. 2) were downwardsweepsin frequencywith time and usuallystartedat a slightlyhigher frequencythan upscreams. Severalstrongharmonicswere evident.Averagefundamentalfrequencies beganat 1495Hz andendedat 1020Hz. The averagedurationwas1011ms.A meanpulserepetitionrateof 886persecondwascalculated fromthe harmonicinterval.Most of the energywasin the second, third,

and fourth

harmonics. Downscreams

(n = 1200)wererecordedfrom 12 whales. 3. Creak

5. Whistle

Whistleswerethe onlyphonations that werenot composedof pUlse-modulated signals(Fig. 5). The frequency rangewashigherthanin theothersoundtypeswitha mean minimumfrequencyof 4268 Hz and a meanmaximumfrequencyof 6608Hz. Theoverallaverage frequency was5000 Hz. Whistledurationaveraged 2.3s,whichwaslongerthan othersoundtypesinvestigated. Sevenanimalsproduced 200 whistles.

6. Tones

This pulse-modulated soundwasdistinguished from upscream, downscreams, andwhinesby muchlessfrequencymodulation (Fig.6).The average fundamental beganand endedat 1344 Hz. The mean modulationfrequencywas 1072Hz. Duration averaged1.5s. Typically,the secondor third harmonicwasstressed. Three-hundred fiftytoneswere

Creakswerecharacterized by a rapidseriesof broad- recorded from seven whales. bandpulseswith energydistributedbetween570 and 7160 Hz, but concentrated between1390and4068Hz (Fig.3). 7. Buzz Creaksaveraged 2.3sin duration.Pulserepetition ratecould Typicalbuzzeswereshort,averaging658 ms (Fig. 7). not be determined from the spectrograms dueto the samThe average energy bandwidth ofbuzzes contained frequenplingrateof theanalysis equipment. Therefore thesemoducies between 1390 and 9136 Hz, but most energy was conlationratesprobablyfell withintheuncertainty windowof the equipment,whichwe estimateto be somewherebetween centratedbetween3000and6150Hz. The highpulserepeti10 and 100 Hz. Creaks were heard most often from Sea tion rate couldnot be determined from the spectrograms. Sevenwhalesproduced350 buzzes. World's animals and were also heard at Sealand and from one animal at Marineworld. A total of 300 creakswere analyzed. 4. Whine

Whinesdid not exhibitthe sweeps in frequency asdid upscreams and downscreams. A slightoverallrise (• = 5

8. Ricochet

This unusualterm for a soundtype resultedfrom its unmistakable resemblance to a ricocheting bullet.To thehumanear,a ricochetgavetheimpression oœbeing a composite of twosounds (Fig.8).Theaverage ricochet contained ener-

FIG. 7. Samplespectrogramof

a buzz(50msperline}.

664

J. Acoust.Sec.Am.,Vol.72, No.3, September 1982

M.E. Dahlheim andF. Awbrey: Vocalization ofcaptivewhales

664

FIG. 8. Sample spectrogram of a ricochet{50 ms per line).

gy between1550and 7647 Hz. Most energyappearedto be

harmonicinterval, was 595 per second.Typically, the sec-

between 2985 and 4970 Hz. Bandwidth was much wider at

ond and third harmonics were stressed. Seesaws were re-

the beginningthan at the end of a ricochet.Durationwas short,averaging703 ms. Two-hundredricochetswere recordedfromfour whales{MarinelandandMarineworld).

corded from three animals at Sea Word, with a total of 150 usable records.

B. Occurrence of two sounds emitted simultaneously 9. Click burst

Click burstswere composedof a rapid and repetitive seriesof pulses(Fig. 9}. The averagedurationof a seriesof clickburstswas 3.8 s. Although usuallyproducedat a high repetitionrate, at timesthis rate wasslowenoughto resolve individual pulses(3-4 per second}.Click burstscontained substantialenergyat frequencies as low as 100 Hz and ex-

tendedabove10kHz. Analaysiswitha frequency windowof 0 Hz to 20 kHz resultedin considerable energyin thesesignalsup to the limitationsof our equipment{19kHz}. Click bunts were recordedfrom eight whaleswhich provideda total of 400 records.

10. Chattot

Chatter alsoconsistedof a rapid seriesof broadband pulses(Fig. 10).The meanminimumandmaximumfrequencieswere340and 9000Hz, respectively. Averageduration was2 s.Pulserepetitionratecouldnotbecalculatedfromthe displays.Three-hundredchatterswere recordedfrom six whales.

On severaloccasions, clicksand frequency-modulated whistlesor amplitude-modulated soundswere emittedsimultaneouslyby oneindividual.Clicksandwhistles,slowed

downeighttimes,aredisplayed in Fig. 12.Two different amplitude-modulated pulsetrainsapparentlycan be producedsimultaneously asevidenced by the sonogram in Fig. 13.

C. Multivariate discriminant analysis

Comparisonof similarsoundtypesrevealedsignificant acousticaldifferencesamongindividuals.How accurately repetitionsof an individual'ssoundswereclassified(by the computer)variedwith soundtype,rangingfrom 100% for creaksto 41.3% for whistles(TableI). This is illustratedin the plotsof the first two discriminantscoresfor individual animalsfor threeof themajorsoundtypes(Figs.14-16).For eachsoundtype,some,but not all, whaleswereperfectly matchedto all repetitionsof their vocalizations (TableII). Whenthecomputermisclassified a voealization froman individual,the misclassified soundwasusuallyplacedwith an animal maintained

11. Seesaw

This shouldsweptrapidly upwardand then downward (Fig. 11).The meanstartingfrequencyof the fundamental was 809 Hz and the mean endingfrequencywas 718 Hz. Averagedurationwas851 ms. Strongharmonicswereevident in seesaws. The meanrepetitionrate, derivedfrom the

at the same oceanarium.

Animalscouldbegroupedacoustically by oceanarium with 47.4% (creak)to 92.7% (ricochet)accuracy,but separationofoceanariumbycreakswasnotstatistically significant,

p > 0.05(TableIII). A plotfor downscreams by oceanarium isshownin Fig. 17.Whena vocalizationfor a specific oceanarium was misclassified,the following pattern emerged.

FIG. 9.Samplespectrogramof a clickburst(50 msper line).

Frequencv (kHz)

665

J. Acoust. Soc. Am., VoL 72, No. 3, September 1982

M.E. Dahlheirnand F. Awbrey:Vocalizationof captive whales

665

FIG. 10. Samplespectrogram of a chatter•50 ms per line).

Marineland and Marineworld tendedto group, as did Sea World and Sealand.VancouverPublic Aquarium showed groupingswith all oceanariums. Ten soundtypessignificantlyseparatedthe sevenmale from the six femalesbut not by their whistles(p = 0.649}. Table IV liststhe accuracyvaluesobtainedfor sexualdiscriminationby soundtypes. III. DISCUSSION

Killer whales,O. orca,producea widerangeof sounds. The limitedvocabulary exhibitedby someof thecaptiveanimalsmayreflectthe amountof timewe spentat a specific oceanarium ratherthantheabsence of thesecalltypesin an individual'svocalrepertoire.In the recordings madeof a newlyborncaptiveanimal,duringthe 15 daysbetweenits birthanddeath,onlysevencatagories of sounds werenoted, ß suggesting thattheyoungof thisspecies mustlearntheother calls(Dahlheimand Moore, in preparation}. Theseseven soundtypeswerealsorecordedfrom the parents.Somecau-

tionshouldbeusedherein interpreting theseresultsbecause

madein Alaskan waters,varientsof thesescreamswere fre-

quentlyheard.Themajorityof calltypeswerecomposed of clicksproduced at variousrepetitionratesand amplitude modulatedand/or frequency modulatedtones. The fact that the trained human ear and statistical tech-

niquescanrecognizeindividualwhalesfromthesoundsthey producemeansthat thesewhalesalmostcertainlycanrecognizeeachother'ssounds. IndividualOrcinuscanapparently be identifiedby most of the soundsthey productbut our analysissuggests that somesoundtypesarebetterpredictors than others.For example,the "whistle"did not appearto differentiate

animals well.

Whendiscriminantanalysismisclassified an individual, the miselassified soundwas usuallyplacedwith an animal maintainedat the sameoceanarium.Althoughwe usedextreme care to note which animal was making a sound,we

mighthaveerredoccasionally. This woulddecrease the apparentaccuracyof the techniquewe usedfor individualdiscrimination.

Greater acousticaldifferenceswere noted among the oceanariumsthan within an oceanarium.Although each

only onenewbornwhalewasrecordedandbehavioral/medi-

oceanarium could be discriminated from the others, the

cal observations indicated that this calf was not normal.

close acoustical resemblances between some of these facili-

The mostprevalentsoundtypesproducedby captive animalswereupscreams and downscreams. In listeningto

ties are extremelysignificant.The geographical origin of capturefor whalesof MarineworldandMarinelandwasthe

tapesmadein the wild by Frank Awbrey (Antarctic),How-

same {British Columbia), as was that of Sea World's and

ardWinn {NorthAtlantic),andfromDahlheim'srecordings

Sealand'sanimals(Lower PugetSound,WA). Vancouver

FIG. 11. Samplesp•ctrogram

ofa seesaw [50msperline)..

0

2

5

6

7

8

9

tO

Frequency (kHz)

666

J. Acoust.Soc. Am., VoL 72, No. 3, September1982

M.E. Dahlheimand F. Awbrey:Vocalizationof captivewhales

666

3•

kHz

FIG. 12Sample sonogram of a whistleand clicks emitted simultaneously.

16 kHz

FIG. 13.Samplesonogram of two amplitude modulated signals emitted simultaneously.

1.2s

Time {seconds)

TABLE I. Discrimination amongindividuals by soundtype. Percent of cases

Soundtype

correctlyclassified

Number of animals

Upscream

58.8

11

Downscream

64.0

12

Creak

100.0

6

Whine

93.1

6

Whistle Tone Buzz Ricochet Chatter Click burst Seesaw

41.3 78.6 88.6 85.4 85.7 65.5 87.9

7 7 7 4 6 8 3

PublicAquariumhadanimalsfrombothregions.Theseresultssuggest dialcoral differences in thecallsofcaptivekiller whales, and a maintenance of these dialects for over ten

years. Dialectal differencescould also help explain the groupingobserved in Fig. 14. Upscreams recordedfrom Marineland/Marineworld animals 6, 7, A, B, and C tended

to cluster.This dialcoralhypothesis is furthersupportedby therecentworkof Ford(1980)whohasrecorded12different podsof free-rangingkiller whalesin the watersof Puget Sound,WA and BritishColumbiaandfounddialectaldifferences..

Sexcouldbe discriminatedby sound.Thesedifferences werenot attributableto call types,but appearedto be more subtlevariations. A differentorderingof theeightacoustical variablesfor sexualseparationwas notedwhenthesewere comparedto theorderingin individualrecognition. All var-

iablescontributed tooveralldiscrimination ofsex,butagain, somesoundtypeswerebettersexpredictorsthan others. This studyshowedthat individual,group,and sexual

11.625

9.250

6.876

5

-2.625

-5.000 -5.000

I -2.625

I -.250

I 2.125

I

I

I

I

4.500

6.875

9,260

11.625

14.00(

Discriminant score 1

FIG. 14. Mean scoresand 95% confidencecircleson the first two discrimi-

nantaxesfor the 11 individuals producing upscreams. (Numbers/letters

o