Scheumannetal.FrontiersinZoology2012,9:36 http://www.frontiersinzoology.com/content/9/1/36 RESEARCH Open Access Vocal correlates of sender-identity and arousal in the isolation calls of domestic kitten (Felis silvestris catus) Marina Scheumann1*, Anna-Elisa Roser1, Wiebke Konerding1, Eva Bleich2, Hans-Jürgen Hedrich2 and Elke Zimmermann1 Abstract Introduction: Human speech does not onlycommunicate linguistic informationbut also paralinguistic features, e.g. information about the identity and the arousal state ofthe sender. Comparable morphological and physiological constraints onvocalproduction inmammals suggest theexistence of commonalities encoding sender-identity and thearousal state of a sender across mammals. Toexplore this hypothesis and to investigate whether specific acoustic parameters encodefor sender-identity whileothers encode for arousal, we studied infants of thedomestic cat(Felis silvestris catus). Kittens are an excellent model for analysing vocal correlates ofsender-identity and arousal. They stronglydependon thecare oftheirmother.Thus, theacoustical conveyance of sender-identity and arousal maybe important for their survival. Results: We recorded calls of 18 kittens inan experimentally-induced separation paradigm, where kittens were spatially separated from their mother and siblings. In theLow arousal condition, infants were justseparated without any manipulation. In theHigh arousal conditioninfants were handled by theexperimenter. Multi-parametric sound analyses revealed that kitten isolation calls are individually distinct and differ between the Low and High arousal conditions. Our results suggested that source- and filter-related parameters are important for encoding sender-identity, whereas time-, source- and tonality-related parameters are important for encoding arousal. Conclusion: Comparable findingsinothermammalian lineagesprovide evidence for commonalities in non-verbal cues encoding sender-identity and arousal across mammals comparableto paralinguistic cues inhumans.This favours the establishmentof general conceptsfor voice recognition and emotionsin humans and animals. Keywords: Affect-intensity, Individual signature, Infant, Mammal, Cat, Vocalisation Introduction (for indexical cues e.g., humans: [3,4], non-human pri- Human speech and non-linguistic vocalisations convey mates: [5,6], Scandentia: [7]; Artiodactyla: [8,9]; paralinguistic cues encoding the physical characteristics Perissodactyla: [10,11]; Carnivora: [12,13]; Cetaceae: of a speaker, termed here indexical cues (e.g., sex, age, [14]; Chiroptera: [15,16]; Rodentia: [17,18]; Proboscidae: body size, sender-identity), and the emotional state of a [19,20]; Sirenia: [21]; Hyracoidea: [22]; for prosodic cues sender, termed here prosodic cues (e.g., emotional see review [23-25]). This suggests a pre-human origin of valence, arousal) (e.g., [1-3]). Whereas linguistic aspects paralinguistic cues due to homologies in the central ner- of humanspeech areunique tohumans,non-verbalcues vous system and the mammalian vocal production comparable to paralinguistic cues were also found in the system. vocalisations of animals of at least 11 mammalian orders In mammals vocal production is based on a highly evolutionarily conserved system. According to the Source-Filter theory of vocal production the respiratory *Correspondence:[email protected] airstream from the lungs passes the larynx (=source) 1InstituteofZoology,UniversityofVeterinaryMedicineHannover,Bünteweg with the vocal folds followed by the supra-laryngeal 17,HannoverD-30559,Germany Fulllistofauthorinformationisavailableattheendofthearticle ©2012Scheumannetal.;licenseeBioMedCentralLtd.ThisisanOpenAccessarticledistributedunderthetermsofthe CreativeCommonsAttributionLicense(http://creativecommons.org/licenses/by/2.0),whichpermitsunrestricteduse, distribution,andreproductioninanymedium,providedtheoriginalworkisproperlycited. Scheumannetal.FrontiersinZoology2012,9:36 Page2of14 http://www.frontiersinzoology.com/content/9/1/36 vocal tract (=filter; [2,26,27]). Indexical cues are sug- weeksafterbirthvisualandauditoryskillsofthekittensas gested to be related to the length, the density and the well as their locomotor and thermoregulatory abilities are tension of the vocal folds (affecting the fundamental fre- limited [52-54] and kittens are completely dependent on quency [28] of the sound signal) and to the length and their mother. Cats have an elaborated vocal repertoire the shape of the supra-laryngeal vocal tract (affecting the [55-59]. Thus, infant vocalisations may play an important formant pattern [29]). Affect-induced physiological role for their survival, signalling their emotional state and changes are suggested to be related to changes in re- theirneeds.Femalesgivebirthtooneto10infantsperlit- spiratory airstream (affecting amplitude, tempo and fun- ter [51]. Litters from different females may be reared in damental frequency [28,30]), changes in muscle tonus the same nest and thus, may become mixed, which could of laryngeal muscles controlling the tension of the vocal make kin signatures essential for offspring recognition folds (causing disruption and changes of fundamental andoffspring-directedmaternalcare[51].Previousstudies frequency [28,30,31]) and changes in the shape, the have already shown that kittens produce isolation calls length and the filter-properties of the supra-laryngeal when isolated from their mother [55,57-60] which evoke vocal tract (affecting formant frequencies [1,30]). maternal behaviour [61]. Context and age-specific varia- Studies in human and non-human mammals demon- tionsintheacousticstructureofkittenisolationcallshave strated that source- and/or filter-related acoustic para- already been described but only for a few acoustic meters are important acoustic parameters encoding parameters [58,60], whereas to our knowledge no data sender-identity (e.g. [4,6]), whereas time-, source- and on acoustically conveyed individual signatures in kitten tonality-related variations are associated with the arousal isolationcallshavebeenpublished. of the sender (e.g.,[23,24,32-34]). Furthermore, non-linear The aim of this study was to investigate the following phenomena (NLP), irregular vibrations of the vocal folds two hypotheses: (1) sender-identity is encoded in the (e.g., subharmonics, biphonations, frequency jumps), have acoustic structure of kitten isolation calls, (2) arousal is become a focal point of acoustic research describing encoded in the acoustic structure of kitten isolation calls highly complex vocalisations (e.g., [35-39]) and are com- and non-linear phenomena occur more often in High mon in human and non-human animals [35-37,39-42]. arousal compared to Low arousal situations. Based on However,theirfunctionisnotyetclear[36,39,43].Onthe these results we aimed to investigate which acoustic one hand, it is argued that NLPs could be important for parameters or sets of acoustic parameters are important individual recognition (e.g., [36,37,39,42]) and on the for encoding sender identity and which are important other hand that NLPs convey information about the for encoding arousal. Vocal correlates of arousal in non- emotional state of the sender (e.g. [37,39]). human animals can be investigated at the behavioural To explorethe impact of certain acoustic parameters on level by measuring different levels of situational urgency encodingsender-identityandarousalinnon-humanmam- within the same behaviouralcontextandlinkingitto the mals, it is important to study both aspects in the same corresponding vocal expression [23]. Thus, we separated individuals using the same set of acoustic parameters and the kittens from their mother and siblings and exposed thesamebehaviouralcontexts.Todate,thereareonlyfew them to two sub-contexts which were assumed to vary in studies investigating both aspects in the same individuals their level of arousal (Low arousal versus High arousal and behavioural contexts (bats: [44,45]; primates: [30,46]; condition). To investigate our hypotheses multi-parametric elephants: [20]; dogs: [47]; tree shrews: [7]) and to our sound analyses were performed measuring 3 time-, 4 knowledge only three studies are available for mammalian source-, 12 filter- and 3 tonality-related parameters infants (elephants: [37]; bats: [45]; cattle: [48]). To explore (Table 1). We will report that a set of source- and filter- the role and potential commonalities of certain acoustic related acoustic parameters is important for encoding parameters or sets of acoustic parameters encoding sender-identity, whereas a set of time-, source- and prosodic and indexical cues in mammalian infant vocali- tonality-related acoustic parameters is important for en- sations, further studies on infants of various mammalian coding arousal. By comparing our findings with data on taxa are needed. other mammals we will explore to which extent our In this study, we explored vocal cues encoding sender- results support the hypothesis for universal acoustic identity(indexicalcues)andarousal(prosodiccues)byin- coding rules expressing indexical and prosodic cues in vestigating infant isolation calls of domestic cats. Cats are mammals due to similar physiological and anatomical an important animal model in human hearing research constraints in the peripheral vocal production system. due to similarities in their auditory system to humans (e.g. [49,50]). Adult females usually live communally in Results small social groups, whereas males live solitarily [51]. We found no significant differences in the acoustic para- Domestic cats are an altricial species, kittens being born metersbetweenindividualswhichwereinitiallyexposedto blind with their ears closed [52]. During the first three the Low or the High arousal condition (Fishers Omnibus Scheumannetal.FrontiersinZoology2012,9:36 Page3of14 http://www.frontiersinzoology.com/content/9/1/36 Table1Descriptionofmeasuredacousticparameters N=16, p≤0.001 except ICI F(15)=1.23, N=16, p=0.256). Parameter Definition For the source- and tonality-related parameters all mea- sured acoustic parameters differed between individuals for Time-relatedparameters both arousal conditions (Low arousal: F(15)≥2.57, N=16, Callduration Timebetweentheonsetandtheoffsetofacall. [ms] p≤0.002;Higharousal:F(17)≥1.96,N=18,p≤0.016).Forthe ICI[ms] Timebetweentheoffsetofacallandtheonsetofthe filter-related parameters almost all measured parameters successivecall. for both arousal conditions differed between indivi- Peaktime[ms] Timebetweentheonsetandthemaximumamplitude duals (High arousal: F(17)≥1.90, N=18, p≤0.022; Low ofacall. arousal: F(15)≥1.85, N=16, p≤0.033 except BWF2 and Source-relatedspectralparameters SD3: F(15)≤1.73, N=16, p≥0.052). To investigate whether MeanF0[Hz] Meanfundamentalfrequencyofacall. calls can correctly classified to the respective individuals, MinF0[Hz] Minimumfundamentalfrequencyofacall. we performed Discriminant function analysis(DFA)com- MaxF0[Hz] Maximumfundamentalfrequencyofacall. bined with Principal Component Analysis (PCA) for each arousalconditionseparately. SDF0[Hz] Standarddeviationofthefundamentalfrequencyofa call. For the Low arousal condition a PCA based on the Filter-relatedspectralfeatures acoustic parameters extracted seven factors (PC) with an Peak[Hz] Frequencywithmaximumenergyoveracall. eigenvalue higher than 1 explaining 71.95% of the variance (seeAdditionalfile1).AnindependentDFAbasedonthese MeanF1[Hz] Meanfrequencyofthefirstformantofacall. sevenPCswasabletoclassify53.13%ofthecallstothere- SDF1[Hz] Standarddeviationofthefirstformantfrequencyofa call. spectiveindividual(cross-validation:41.88%)whichwassig- nificantly above chance level (6%; p<0.001). On an BWF1[Hz] Bandwidthofthefirstformantfrequencyofacall. individual level for 15 out of 16 subjects for the original MeanF2[Hz] Meanfrequencyofthesecondformantofacall. classification and for 12 out of 16 subjects for the cross- SDF2[Hz] Standarddeviationofthesecondformantfrequencyof acall. validation significantly more calls were correctly classified than expected by chance (p≤0.019). The DFA calculated BWF2[Hz] Bandwidthofthesecondformantfrequencyofacall. seven DFs. Thereby, DF1, 2 and 3 explained 86.6% of the MeanF3[Hz] Meanfrequencyofthethirdformantofacall. variationinthecalls.DF1showedthehighestcorrelationto SDF3[Hz] Standarddeviationofthethirdformantfrequencyofa call. PC1 (r=0.568), DF2 showed the highest correlation to PC6 (r=0.698), whereas DF3 showed the highest correlation to BWF3[Hz] Bandwidthofthethirdformantfrequencyofacall. PC2(r=−0.593).PC1showedthehighestloading factorsto F2–F1[Hz] Differencebetweenthemeanofthesecondandthe firstformantfrequency. thesource-relatedparameters:MeanF0,MinF0andMaxF0 (r≥0.751;Table2)andto thefilter-relatedparameter F2-F1 Consistency Meanmaximumcorrelationofpowerspectraof successive25mstimestepsofacall. (r=−0.704). PC2 showed the highest correlation to the Tonality-relatedparameters filter-relatedparameters:MeanF1andSDF1(r≥0.755).PC6 Cepstralpeak Valueofthepeakatthefundamentalperiodofa showednoloadingfactorsabove0.700. [V] cepstrumforthemiddle10msofthecall. For the High arousal condition a PCA based on the Voiced[%] Percentageofvoicedframesofacall. acoustic parameters extracted seven factors with an MaxHNR[db] Maximumharmonic-to-noiseratioofacall. eigenvaluehigherthan1explaining68.90%ofthevariance (see Additional file 1). An independent DFA based on thesesevenPCswasabletoclassify63.33%ofthe calls to test: χ2≤55.55, df=44, p≥0.114 for both conditions). This the respective individual (cross-validation: 47.78%) which suggeststhattheorderinwhichthesubjectswereexposed was significantly above chance level (6%; binomial test: to the two arousal conditions did not affect the acoustic p<0.001).Onanindividuallevelforallsubjectsfortheori- parameters of their vocalizations. Therefore, both groups ginal classification and for 16 out of 18 subjects for the werepooledforfurtheranalysis. cross-validation significantly more calls were correctly classified than expected by chance (p≤0.019). The DFA Sender-identity calculated seven DFs. Thereby, DF1, 2 and 3 explained Forbotharousalconditionsthemajorityoftime-,source-, 82.9%ofthevariationinthecalls.DF1showedthehighest filter- and tonality-related parameters showed significant correlation to PC1 (r=−0.730), DF2 showed the highest differences between individuals (Fisher Omnibus test: correlation to PC2 (r=0.700), whereas DF3 showed the χ2≥784.64,df=44,p<0.001;Table2).Fortime-relatedpara- highest correlation to PC3 (r=0.706). PC1 showed the metersalmostallparametersdifferedsignificantlybetween highest loading factor to the filter-related parameters: individuals for both arousal conditions (High arousal: Peak, MeanF2, F2-F1 (r≥0.711; Table 2). PC2 showed the F(17)≥1.89, N=18, p≤0.022; Low arousal: F(15)≥2.69, highest loading factor to source-related parameters: Scheumannetal.FrontiersinZoology2012,9:36 Page4of14 http://www.frontiersinzoology.com/content/9/1/36 Table2Resultsoftheone-wayAnovatestingfordifferencesbetweenindividualsforeachacousticparameterand arousalconditionandthecorrelationcoefficientwiththethreemostimportantPCsfortheDFA;LOW=Lowarousal condition;HIGH=Higharousalcondition;boldp-valuesrepresentsignificantdifferencep<0.05;boldloadingfactors representtheparametersshowingloadingfactorshigherthan0.700withtherespectivePC LOW HIGH Parameters F p PC1 PC2 PC6 F p PC1 PC2 PC3 Time-relatedparameters Callduration[ms] 6.632 <.001 -.368 .365 .324 5.574 <.001 -.147 -.375 .132 ICI[ms] 1.230 0.256 -.080 -.113 -.218 1.894 0.022 .255 .168 .124 Peaktime[ms] 2.688 0.001 -.203 .056 .487 3.978 <.001 -.032 -.412 .145 Source-relatedparameters MeanF0[Hz] 25.331 <.001 .863 .047 .304 20.199 <.001 -.380 .810 .018 MinF0[Hz] 16.034 <.001 .864 -.149 .050 10.574 <.001 -.400 .650 -.312 MaxF0[Hz] 27.394 <.001 .751 .213 .426 17.166 <.001 -.341 .818 .166 SDF0[Hz] 2.806 0.001 -.361 .363 .380 5.921 <.001 .012 .371 .602 Filter-relatedparameters Peak[Hz] 3.919 <.001 .387 .255 .132 10.444 <.001 -.725 -.236 .071 MeanF1[Hz] 8.305 <.001 .207 .814 -.081 8.631 <.001 -.677 -.202 .403 SDF1[Hz] 3.170 <.001 -.122 .755 .048 4.558 <.001 -.047 .172 .614 BWF1[Hz] 1.848 .033 -.015 .582 -.311 1.953 .017 .207 .112 .486 MeanF2[Hz] 4.287 <.001 -.668 .078 .326 11.260 <.001 .711 .315 .174 SDF2[Hz] 2.322 .005 -.147 .289 -.080 4.707 <.001 .406 -.133 .414 BWF2[Hz] 1.130 .335 -.065 -.146 .072 1.896 .022 .281 -.143 .031 MeanF3[Hz] 3.411 <.001 -.538 .022 .047 5.251 <.001 .658 .170 .046 SDF3[Hz] 1.727 .052 -.247 .306 .102 2.442 .002 .107 -.451 .455 BWF3[Hz] 2.086 .014 -.152 -.363 .047 2.386 .003 -.031 -.313 .045 F2-F1[Hz] 6.789 <.001 -.704 -.385 .333 14.675 <.001 .850 .329 -.072 Consistency 2.072 0.014 .205 -.474 .195 5.135 <.001 .140 -.110 -.695 Tonality-relatedparameters Cepstralpeak[V] 3.902 <.001 .174 .632 .038 3.501 <.001 -.230 .074 .492 Voiced[%] 2.569 0.002 .459 -.061 .090 1.963 0.016 .091 .351 .025 MaxHNR[db] 4.058 <.001 .556 -.197 .171 2.174 0.007 -.438 .169 -.156 MeanF0 and MaxF0 (r≥0.810). PC3 showed no loading per litter so that one litter can contribute more to the factorabove0.700toanyoftheacousticparameters. resultsthananother. Comparing the classification accuracy between both Wefoundalmostnosignificantdifferencesintheacous- arousal conditions showed no significant differences tic parameters between sexes and almost no significant (original: t(15)=1.29, N=16, p=0.215; cross-validation; correlations with body weight. For the factor sex in the t(15)=0.426, N=16, p=0.676) demonstrating that the High arousal condition only the BWF3 and in the Low level of individual distinctiveness was similar for both arousalconditiononlytheSDF2andSDF3differedsignifi- arousal conditions. Performing a crossed pDFA inves- cantly between sexes (t(16)≥|2.45|, N=18, p≤0.026). For tigating differences between subjects by controlling the factor body weight a significant negative correlation for the arousal level also revealed that individuals withcalldurationandasignificantpositivecorrelationfor could significantly correctly be classified (original: the percentage of voiced frames was found only for the p=0.004; cross-validation: p=0.002). Low arousal condition (r≥|0.540|, N=18, p≤0.021). How- Performing a nested pDFA testing for differences ever, controlling for multiple testing, using the Fisher between subjects by controlling for litter confirmed Omnibus test, showed that these differences could be significant differences between individuals (original and explainedbychance(sex:χ2=104.33,df=88,p=0.113;body cross-validation:p≤0.001forbotharousalconditions).This weight:χ2=101.09,df=88,p=0.161).Thisindicatesthatin- suggests that individual differences cannot be explained dividual differences cannot be explained by sex or body by the fact that we used a varying number of kittens weight. Furthermore, the body weight of kittens did Scheumannetal.FrontiersinZoology2012,9:36 Page5of14 http://www.frontiersinzoology.com/content/9/1/36 not differ between sexes (t(16)=1.09, N =N =9, in the Low arousal condition, whereas ICI was shorter female male p=0.292). (t(17)≥|2.58|, N=18, p≤0.019; Figure 1). Peaktime sho- All in all, almost all measured acoustic parameters dif- wed a tendency to be longer in the High than in the fered between individuals for both arousal conditions. Low arousal condition (t(17)=−1.92, N=18, p=0.072). However, classificationof individuals was mainlyattribu- For the source-related parameters three out of four ted to source- and filter-relatedparameters. Thereby,the parameters differed significantly between conditions. parameters which seem to be most important for classi- Thus, the MeanF0, MinF0 and MaxF0 were lower in fication were similar across conditions, suggesting con- the High compared to the Low arousal condition sistencyacrossdifferentarousallevels. (t(17)≥3.12, N=18, p≤0.006; Figure 1). For the filter- related parameters six out of 12 parameters diffe- Arousal red significantly between conditions. Thus, Peak and Arousal affects acoustic structure of kitten isolation calls MeanF1 were higher in the High than in the Low for time-, source-, filter- and tonality-related parameters arousal condition, whereas SDF1, BWF1 and F2-F1 (Fisher Omnibus test: χ2=175.55, df=44, p<0.001; were lower in the High versus the Low arousal condi- Table 3). For time-related parameters two out of three tion (t(17)≥|2.13|, N=18, p≤0.048). Furthermore, the parameters differed significantly between arousal condi- consistency was lower in the High compared to the tions.Thereby,call duration waslongerintheHigh than Low arousal condition (t(17)=3.03, N=18, p=0.008). For Table3MeanandstandarddeviationoftheacousticparametersforLowandHigharousalcondition,resultsofthe dependentt-testcomparingbotharousal-levelsforeachacousticparameterandthecorrelationcoefficientwiththe PC1;boldp-valuesrepresentsignificantdifference;↑valueishigherintheHighthanintheLowarousalcondition, ↓valueislowerintheHighthanintheLowarousalcondition;boldloadingfactorsrepresenttheparametersshowing loadingfactorshigherthan0.700withtherespectivePC LOW HIGH LOWversusHIGH Parameters Mean SD Mean SD T p PC1 Time-relatedparameters Callduration[ms] 566.34 168.62 707.10 186.09 −2.81 .012 ↑ -.756 ICI[ms] 2072.53 1442.76 1075.38 652.32 2.58 .019 ↓ .482 Peaktime[ms] 0.23 0.08 0.29 0.12 −1.92 .072 -.640 Source-relatedparameters MeanF0[Hz] 1305.42 238.49 1105.10 184.60 3.82 .001 ↓ .746 MinF0[Hz] 931.71 274.86 746.65 166.58 3.12 .006 ↓ .785 MaxF0[Hz] 1517.21 249.64 1316.52 221.48 3.68 .002 ↓ .686 SDF0[Hz] 154.99 37.51 149.56 37.23 .50 .623 -.077 Filter-relatedparameters Peak[Hz] 1648.68 327.77 2493.48 676.96 −5.24 <.001 ↑ -.610 MeanF1[Hz] 2112.80 420.47 2642.38 325.62 −4.84 <.001 ↑ -.509 SDF1[Hz] 696.02 273.16 549.43 159.78 2.13 .048 ↓ .107 BWF1[Hz] 1120.22 466.85 623.80 376.95 3.60 .002 ↓ .442 MeanF2[Hz] 7034.30 570.61 6758.39 511.89 1.79 .091 .095 SDF2[Hz] 981.97 286.65 987.88 297.99 -.07 .948 -.247 BWF2[Hz] 1977.26 512.96 1858.01 732.96 .75 .463 -.067 MeanF3[Hz] 11320.63 552.86 11240.16 524.16 .47 .642 .118 SDF3[Hz] 1134.20 249.39 1273.83 231.80 −1.52 .148 -.588 BWF3[Hz] 2044.74 1128.99 3017.33 1602.68 −2.03 .058 -.410 F2-F1[Hz] 4921.50 694.44 4116.01 745.54 4.31 <.001 ↓ .348 Consistency 0.89 0.02 0.86 0.03 3.03 .008 ↓ .312 Tonality-relatedparameters Cepstralpeak[V] 2.36 0.59 2.69 0.61 −1.69 .110 -.366 Voiced[%] 98.23 1.71 96.26 2.67 2.53 .022 ↓ .712 MaxHNR[db] 31.73 4.61 28.78 3.27 2.51 .022 ↓ .576 Scheumannetal.FrontiersinZoology2012,9:36 Page6of14 http://www.frontiersinzoology.com/content/9/1/36 Figure1MeanandstandarddeviationfortheLowandHigharousalconditionfortheacousticparameterofkittenisolationcalls whichhadimportantimpactontheclassificationofarousal;t(17)≥|2.53|,N=18,p≤0.022. the tonality-related parameters two out of three para- containing NLPs was not significantly different between meters differed significantly between arousal conditions. theLowandtheHigharousalcondition(mean =50.00%; Low Thus, the percentage of voiced frames and MaxHNR mean =45.00%; Z=−0.358, n=17, N=18, p=0.720). The High were lower in the High compared to the Low arousal most often seen NLP was chaos (33.61%, N=18), followed condition(t(17)≥|2.51|,N=18,p=0.022;Figure1). by frequency jumps (15.43%, N=14) and subharmonics Based on the means of the acoustic parameters for (9.26%, N=8). We found no significant differences in the each individual and arousal condition a PCA extracted percentage of calls containing frequency jumps (mean- six factors with an eigenvalue higher than 1 explaining =20.00%; mean =10.56%; Z=−1.84, n=12, N=18, Low High 81.28% of the variance (see Additional file 1). An inde- p=0.066) or chaos (mean =38.89%; mean =28.89%; Low High pendent DFA based on these six PCs was able to assign Z=−1.03, n=15, N=18, p=0.304) between the Low and the 88.9% of the cases to the respective arousal condition High arousal condition. In contrast, subharmonics were (cross-validation: 80.06%), which was significantly above onlyobservedintheHighandnotintheLowarousalcon- chance level (50%; for original and cross-validated dition (mean =0.00%; mean =18.33%; Z=−2.55, n=8, Low High classification: both conditions: binomial test: p<0.001; N=18,p=0.011). Low arousal: p=0.008; High arousal: p=0.031; Figure 2). Altogether,arousalconditionsdifferedintime-,source-, Thereby, PC1 showed the highest correlation with the filter- and tonality-related parameters. However, for discriminant function (r=0.709), whereas the other classification the most loading acoustic parameters factors showed correlations lower than |0.219|. PC1 were call duration, percentage of voiced frames, mean showed the highest loading factors to call duration and minimum fundamental frequency. In the High (r=-0.756), MinF0 (r=0.785), MeanF0 (r=0.746) and arousal condition significantly more calls containing percentage of voiced frames (r=0.712; Figure 1). subharmonics could be observed, whereas the occur- Analysing non-linear phenomena we detected NLPs in rence of other NLPs did not differ between the two 47.46% of the analysed calls, but the percentage of calls arousal conditions. Scheumannetal.FrontiersinZoology2012,9:36 Page7of14 http://www.frontiersinzoology.com/content/9/1/36 in the acoustic structure of kitten isolation calls can be perceived by the mother, since Härtle [55] demonstrated that mothers recognise their kittens from their voices. Thus,individualsignaturesininfantisolationcallswould allow the mother to discriminate their own infant from those of others, to direct their care-giving behaviour and thereby increase their own fitness. This suggests that these individual signatures in kitten isolation calls may beanimportanttoolfor kinselection. We found no effect of sex on the acoustic structure of kitten isolation calls, this being in agreement with other studies on small-bodied animals (e.g., tree shrews: [7]; pygmy marmosets: [65]; bats: [66]), whereas the majority of studies on large-bodied animals revealed sex-specific differences (see review on primates: [67]). Ey and collea- gues [67] argued that these sex-specific differences were Figure2ScatterplotforthePC1andPC2ofthearousal mainly driven by differences in body size due to sexual analysis. dimorphism. Since the kittens at this age did not show such a sex dimorphism in body weight, no differences in Discussion the acoustic structure of kitten isolation calls was The results clearly show that in kitten isolation calls expected. We also found no influence of body weight, sender-identity and arousal-level are encoded by diffe- which is also in agreement with findings of other studies rent combinations of acoustic parameters. Although uni- (e.g., see review on primates [67] and additionally tree variate analysis showed that almost all kinds of acoustic shrews: [7]). Ey and colleagues [67] argued that a rela- parameters varied between sender-identity and arousal, tionship between body size and acoustic parameters is DFA combined with PCA suggested that the impact of highly predictable when body size variation is large but certain parameters differed. Sender-identity was mainly less predictable if variation is small. Thus, it could be determined by a combination of source- and filter- argued that the variation in body weight is not large related parameters, whereas arousal level was mainly enough to affect acoustic structures of vocalisations in determined by a combination of time-, source- and kitten isolation calls (mean=307.33 g; range: 246–370 g; tonality-relatedparameters. SD=33.03). All in all, kitten isolation calls contain indi- vidual signatures, which cannot be explained by sex or Sender-identity body weight. Kitten isolation calls differed between individuals in almost all acoustic parameters independent of arousal Arousal condition and could correctly be classified above Our hypothesis that arousal is encoded in acoustic para- chance level, supporting our hypothesis that sender- meters of kitten isolation calls was supported. Calls identity is encoded in the acoustic structure of kitten recordedintheHigharousalconditionwerecharacterised isolation calls. Analysis showed that this cannot be bylongercallduration,ashorterintercall-interval,alower explained by the fact that we used a varying number fundamental frequency, a higher peak- and first formant of kittens per litter so that one litter can contribute frequencyand lowertonalityvaluesthancallsrecordedin more to the results than another. Thus, the pDFA the Low arousal condition. This is partly in agreement controlling for litter also revealed differences in the with other studies in cats investigating whether acoustic acoustic structure between kittens. structureofisolationcallsvariesbetweencontexts[58,60]. Individual distinctiveness was found for both arousal Our results are in line with the finding of Haskins [60] conditions and could also be approved by pooling both andRomandandEhret[58]thatcalldurationwasshorter conditions using a pDFA. Thereby, for both arousal con- in low arousal contexts (Isolation without manipulation) ditions almost the same source- and filter-related para- than in high arousal context comparable to our High meters (MeanF0, MaxF0, F2-F1) contributed mainly to arousalcondition(namelyaRestraincontext[60],Picked- the classification result. This suggests that individual dif- up and Tail-pressing context [58]). Regarding our finding ferences are consistent across different arousal levels. thatthefundamentalfrequencywasdecreasedintheHigh This is in agreement with several studies showing arousal condition in comparison to the Low arousal that infant isolation calls contain individual signatures condition, our data are not in agreement with those of (e.g., [16,62-64]). It can be assumed that these variations Haskins [60] who found no significant differences in the Scheumannetal.FrontiersinZoology2012,9:36 Page8of14 http://www.frontiersinzoology.com/content/9/1/36 fundamental frequency between the Isolation and the animals (e.g., [7,20,37]). The decrease in tonality may go Restrain context. However, Romand and Ehret [58] found along with an increase in non-linear phenomena due to a thatthefundamentalfrequencybecamesignificantlylower loss of vocal control [37]. However, we found only a diffe- in the Tail-pressing context than in the Isolation context rence in the percentage of calls containing subharmonics whenkittensturned32daysold. between the Low- and the High arousal condition but not Comparingourresultswithotheranimaltaxawefound forNLPsingeneral,chaosorfrequencyjumps.Stoegerand that for the temporal parameters similar changes are colleagues [37] found a positive correlation between reportedfor avariety of mammaliantaxaand behavioural harmonic-to–noise ratio (HNR) and duration of chaotic contexts (see review [23,24]). Concerning source-related segments. Since we found a decrease in the MaxHNR it parameters the results are controversial. Thus, the majo- couldbeassumedthatalthoughtheoccurrence(percentage rity of studies found either an increase of fundamental of calls) is the same the relation of NLP in the call differs. frequency with increasing arousal or no effect (see review In the data set we used for these analyses we could not [23,24]).Surprisingly,wefoundadecreaseinfundamental alwaysdecidereliablywhenachaoticcomponentstartedor frequency from Low to High arousal condition. As finished.Therefore,furtherstudiesareneededtoinvestigate describedabove,alsoRomandandEhret [58] found a de- the role and function of non-linear phenomena in kitten crease in F0 from the Tail-pressing context (similar to isolationcalls. our High arousal condition) compared to the Isolation To expose animals to a situation assumed to induce a context (similar to our Low arousal condition) in 32–46 specific emotion and measuring the corresponding beha- day-old kittens. Furthermore, during male-male inter- vioural and physiological changes is a general approach in action it was shown for grey mouse lemurs that the animalemotionalresearch[24].Vocalcorrelatesofarousal start fundamental frequency of their calls was lower in wereinvestigatedbyexposingsubjectstodifferentlevelsof contexts where they had physical fights (assumed to re- situational urgency within the same behavioural context flect high arousal) compared to contexts where they and analysing the acoustic parameters of their vocal had no physical contact (assumed to reflect low arousal expressions(e.g.,[7,23,30,34,44]).Inthisstudykittenswere in the animal) [68]. separated from their mother and siblings in both condi- For the filter-related parameters we found an increase of tions. In the Low arousal condition they were left undis- the peak frequency and the frequency of the first formant turbed whereas in the High arousal condition they were fromLowtoHigharousalcondition.Anincreaseinthefre- additionally manipulated by the experimenter assumed to quency of filter-related parameters was also found for pigs induce a higher level of urgency/arousal. However, al- [69], primates [30,70] and tree shrews [7]. An increase in though if we assume that the general behavioural context the frequency of the first formant (=resonance frequency) and the emotional quality might be fairly similar between wasalsofoundinpigs[71]andchimpanzees[70].Further- the sub-contexts, we can not rule out that the meaning/ more, adecrease in the consistency agrees with findings in function of vocalizations differs between sub-contexts. To tree shrews [7]. The increase in peak frequency and for- clarifythis,furtherstudiesareneededwhichexposekittens mant frequencies could be explained by the extent of todifferentcontextsassumedtovaryinarousalandalsoin mouthopeningwhichresultsinashortervocaltractlength emotionalqualityandcomparetheirresponses. [72]. It could be argued that the changes we found for the All in all, we found that arousal-related changes of acoustic parameters, especially those of filter-related para- time- and tonality-related parameters in kitten isola- meters,couldbeattributedtothemanipulationintheHigh tion calls correspond with previous findings in other arousal condition. This means by turning the kittens on mammalian taxa. their back the length of the vocal tract may be changed. However, we did not systematically manipulate the head Conclusion positionso that the angle between the head and the breast In conclusion, our results showed that kitten isolation couldvarybetweenkittens.Duetothisunsystematicalvari- calls encode sender-identity and arousal. Thereby, diffe- ationofheadposition,itwouldbeunlikelythattheanalysis rentsetsofparametersseemtobeimportant.Thus,time-, of sender-identity favoured the same source- and filter- source- and filter-related parameters mainly encode for related parameters for both arousal conditions. Thus, we arousal, whereas source- and filter-related parameters suggestthatturningthekittenontoitsbackcannotaccount mainly encode for sender-identity. Thereby, source- for the increase in filter-related parameters. Instead, we related parameters seem to be important for both coding favour the assumption that mouth opening shortens the the sender-identity and arousal. This suggests that based vocal tract, resulting in an increase of filter-related para- on parameters of the fundamental frequency alone we meters which was already shown for cats by Shipley and cannot differentiate between sender-identity and arousal. colleagues [72]. The decrease in tonality from Low arousal Instead, we argue that single parameters alone do not to High arousal condition agrees with findings in other codeforarousalandsender-identity(especiallybecauseall Scheumannetal.FrontiersinZoology2012,9:36 Page9of14 http://www.frontiersinzoology.com/content/9/1/36 vary) but that certain sets or relations of parameters struggledwiththeirlegsandtriedtoturnaround.Thus,we encode sender-identity or arousal. Thus, playback studies assume that the strong manipulation by the experimenter are needed, manipulating specific acoustic parameters, to in the High arousal condition induced a higher level of verifywhichacousticparametersarebiologicallyimportant urgency/arousalinthekittencomparedtotheLowarousal forrecognisingsender-identityandarousal. conditionwheretheywereleftundisturbed. Kittens were tested in one session. In this session both Material & methods conditions were performed in a randomised order for 3 Subjectsandhousing minutes each. After finishing a condition kittens were Wetested18mongrelkittens (9males, 9 females) from6 reunited with their mother and siblings before the other litters aged 9 to 11 days and housed in the SPF (Specific condition was performed. The inter-condition interval Pathogen Free) breeding colony at the Hannover Medical was dependent from the number of siblings. Thus, we School.Allkittenswererearedbytheirmothers.Theani- tested the kittens of one litter one after another in the mal husbandry there complies withthe recommendations first condition. After finishing this test for all kittens we for domestic cats noted in Appendix A of the European started to test the kittens in the same order for the ConventionfortheProtectionofVertebrateAnimalsused second condition. To avoid stress for the mother, the for Experimental and other Scientific Purposes (ETS mother remained in the animal room but was prevented No.123) (http://conventions.coe.int/Treaty/EN/Treaties/ from coming into contact with the kittens during the PDF/123-Arev.pdf). One mother and her kittens lived in experimental trial by the animal keeper (e.g., groomed one animal room (12.5 m2 to 20.6 m2) equipped with a orplayed withthemother). wooden nest box, an infrared lamp as additional heat Kitten vocal responses were recorded using a source, bars for scratching and plastic items for playing. Sennheiser microphone (ME67,Sennheiser,Wedemark, Cats were used to the daily routine of animal keepers Germany; frequency range: 40 – 20,000 Hz) linked to a entering the animal rooms and playing with or grooming Marantz professional solid state recorder (PMD 660, them. All kittens were familiar with being handled by Marantz, Osnabrück, Germany; sampling frequency: humans due to the daily weighing routine and mothers 44.1 kHz, 16 bit). Sound files were stored as wave were used to the kittens being removed for a short time files on a Compact Flash memory card (4 GB, Scan from the nest box. Furthermore, they had acoustic and Disk Corporation, Milpitas, CA, USA). The kittens’ olfactory contact to other cats. The mother was fed daily behaviour were videotaped using a digital camcorder with canned (Pet, De Haan Petfood, Nieuwkoop, the (Sony DR-TRV 22E-PAL, Tokyo, Japan). Netherlands) and dry cat food (SDS Pet Food, Special Diets Services, Witham, Essex, UK). Additionally, freshly Acousticanalysis killed rats were provided daily together with milk or curd Vocal recordings were visually inspected using spectro- cheese. Water was available ad libitum. Animals were grams of the software Batsound PRO 3.31 (Pettersson housed at a temperature of 22±2°C, relative humidity of Elektronik AB, Uppsala, Sweden). Isolation calls were 55±5% and a light/dark cycle of 12:12 hours (lights on at characterised as tonal calls with a rise and fall in the 6:00a.m.). fundamental frequency with peak intensity around the mid-point (Figure 3a; [57]). For each individual and each Experimentalprocedureanddatarecording arousal condition we selected 10 calls of good quality Experiments were performed in the animal rooms of the with a minimum amplitude difference of 5% between respective mother and her kittens. We conducted a sepa- background noise and maximum amplitude of the call. rationparadigminwhicheachkittenwasremovedfromits For the Low arousal condition we selected the first 10 nestbox and spatially separated from its mother and sib- calls of good quality. For the High arousal condition we lings. To induce two different levels of arousal in a kitten selected the first 10 calls of good quality after turning (theLowandHigharousalcondition),kittenswereexposed the kitten onto its back (except for one kitten which was to two sub-contexts varying in the level of situational ur- only lifted up so that its legs had no contact to the gency.Thus,intheLowarousalconditionakittenwasonly ground). In total, we analysed 348 calls from 18 indivi- spatiallyseparatedfromitsmotherandsiblingsandleftun- duals. For two individuals only three and five calls were disturbed by the experimenter (=placed alone on the floor availableintheLowarousalcondition. oftheanimalroom),whereasintheHigharousalcondition We performed a multi-parametric sound analysis a kitten was additionally manipulated by the experimenter using the software Batsound PRO 3.31, SIGNAL 3.1 i.e. the kitten was grasped, lifted off the ground and/or (Engineering Design, Berkeley, California, U.S.A.) and turned onto itsback so that thelegs had no contact to the PRAAT (www.praat.org; [73]) combined with GSU ground. In the Low arousal condition kittens moved PRAAT TOOLS [74]. The software Batsound PRO around slowly, whereas in the High arousal condition they was used to manually measure the call duration and Scheumannetal.FrontiersinZoology2012,9:36 Page10of14 http://www.frontiersinzoology.com/content/9/1/36 Figure3Exampleofkittenisolationcalls;(a)harmonicisolationcallwithoutnon-linearphenomena,(b)isolationcallwithafrequency jumpandachaoticcomponent,(c)isolationcallwithsubharmonics.