Spectrograms of examples of the four call types. Sampling frequency for all calls was 48 kHz with 50% overlap. Window size is 100 ms for trumpet, roar, and chirp, and 200 ms for rumbles. Note the different Y -axis scale in G–I. 

Contexts in source publication

Context 1

... for most other calls. Spectral envelope analysis was carried out using linear predictive coding ͑ LPC ͒ . Calls other than rumbles were re-sampled at 16 000 Hz whereas rumbles were re- sampled at 600 Hz. LPC analysis was carried out using the autocorrelation method with a time step of 0.005 s and a window length of 0.005 s ͑ 0.05 s for rumbles ͒ . Harmonicity ͑ harmonics-to-noise ratio ͒ was analyzed using the cross- correlation algorithm with a time step of 0.05 s except for chirps, for which 0.005 s was used due to the short duration of the calls. Calls of different age-sex classes adult male, adult female, young male, and young female ͒ were analyzed sepa- rately. The measured features include the call frequency range, duration, harmonicity, as well as mean, minimum, and maximum values of the fundamental frequency ͑ F 0 ͒ ͑ Table I ͒ . In addition, the frequencies and amplitudes of the LPC peaks were measured to examine spectral patterning. The analyzed acoustic features were separated according to the four age-sex classes and compared statistically ͑ for all categories where the sample size was Ն 3 ͒ for each call type using a combination of one-way analysis of variance and pair-wise comparisons of means by unpaired t -tests. The fea- tures compared included mean F 0 , minimum F 0 , maximum A total of 371 calls were recorded from 154 individuals. F 0 , call duration, harmonicity, and frequency and amplitude Of these, 258 calls were analyzed from 109 individuals, con- of the peaks of the LPC spectrum. sisting of 14 males and 95 females ͑ Table II ͒ . The remaining For the rumbles, the start and end frequencies of F 0 , and calls were discarded on account of low signal-to-noise ratio percentage time from the start to the maximum and mini- or overlap with other calls due to simultaneously vocalizing mum frequencies were measured ͑ Table I ͒ . To characterize individuals. Based on structural characteristics, the calls the frequency modulation in finer detail two measures were could be classified into four types, namely, trumpets, chirps, used ͑ Table I ͒ . The fourth harmonic of all calls was used roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, since visual inspection of the spectrograms revealed that the could be distinguished by its unique frequency range ͑ 10– modulation could be measured at high resolution across al- 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the most all recordings whereas the fundamental was sometimes other calls. Chirps were distinguished by their unique tem- contaminated with noise. The higher harmonics may also poral structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of have greater functional relevance in social recognition as ar- much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. gued by McComb et al. ͑ 2003 ͒ based on playback experi- Trumpets and roars differed from chirps in their duration ments on wild African elephants. The first measure was the ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their number of observed peaks in the fourth harmonic of each spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly call, determined visually from the spectrogram with a mini- lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . mum frequency modulation of 4 Hz as a cutoff. The second There were no significant differences between the different age-sex classes in all of the call features that were exam- measure was the frequency change per unit time for the ined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, fourth harmonic, measured by the cumulative change in fre- wherein females had significantly lower mean F 0 and mini- quency divided by the time of the call ͑ window length of 200 mum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, ms ͒ . t = 2.78 ͒ than males ͑ Table III ͒ . To detect the presence of structural subtypes within the rumbles, 13 measured call features ͑ except harmonicity and LPC peaks ͒ were used to generate pair-wise Euclidean distances between calls. The distance matrix was then subjected to cluster analysis using the Unweighted Pair Group Method with Arithmetic mean ͑ UPGMA ͒ ͑ Sneath and Sokal, 1973; Manly, 1986 ͒ . All statistical analyses were performed using STATISTICA ͑ 1999, Statsoft Inc., Tulsa, USA ͒ . A total of 371 calls were recorded from 154 individuals. Of these, 258 calls were analyzed from 109 individuals, con- sisting of 14 males and 95 females ͑ Table II ͒ . The remaining calls were discarded on account of low signal-to-noise ratio or overlap with other calls due to simultaneously vocalizing individuals. Based on structural characteristics, the calls could be classified into four types, namely, trumpets, chirps, roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, could be distinguished by its unique frequency range ͑ 10– 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the other calls. Chirps were distinguished by their unique temporal structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. Trumpets and roars differed from chirps in their duration ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . There were no significant differences between the different age-sex classes in all of the call features that were examined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, wherein females had significantly lower mean F 0 and minimum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, t = 2.78 ͒ than males ͑ Table III ͒ . The different call types were associated with characteristic body postures and behaviors of the vocalizing individual and other members of the herd. These are described in detail in Table IV. Trumpets were loud, conspicuous high-frequency calls. They were in the frequency range of 405–5879 Hz with a mean duration of about 1 s ͑ Table V ͒ . They had a rich harmonic structure with at least seven clearly visible harmonics ͑ Fig. 1, A–C ͒ . Spectral envelope analysis revealed the first frequency peak to be at 706 Hz ͑ Fig. 2 ͒ . The fourth frequency peak ͑ at 3078 Hz ͒ was about 13 dB lower in amplitude than the first frequency. Out of 73 trumpets where the age-sex class was unambiguous, 58 ͑ 79.5% ͒ were produced by adult or sub-adult females, nine ͑ 12.3% ͒ by juvenile females, five ͑ 6.8% ͒ by adult or sub-adult males, and one 1.4% by a juvenile male ͑ Table II ͒ . Out of the 71 trumpets where the context was clear, seven ͑ 9.9% ͒ were in the context of play, 17 ͑ 23.9% ͒ in the context of disturbance by humans or vehicles, 29 ͑ 40.8% ͒ in the context of disturbance by other non-human species, ten ͑ 14% ͒ in the context of inter-specific aggression, and eight ͑ 11.2% ͒ while running out of a waterhole ͑ Fig. 3 ͒ . Spe- cifically, trumpeting was observed while encountering other species such as deer ͑ Axis axis ͒ , gaur ͑ Bos gaurus ͒ , dhole ͑ Cuon alpinus ͒ , bears ͑ Melursus ursinus ͒ , tigers ͑ Panthera tigris ͒ , egrets ͑ Egretta garzetta ͒ , and humans. Roars were noisy, long calls that were in the frequency range of 305–6150 Hz and had a mean duration of about 2 s ͑ Table V, Fig. 1, D–F ͒ . They were in the same frequency range as trumpets and their spectral patterning was also simi- lar with seven frequency peaks Fig. 2 , the first frequency peak being at 656 Hz. The amplitude fell steeply with increasing frequency, the fourth frequency peak being 23 dB below the first ͑ Fig. 2 ͒ . This was in contrast to trumpets, where the fourth frequency peak was about 13 dB below the first ͑ Fig. 2 ͒ . Roars also showed significantly lower harmonicity than trumpets ͑ Table III ͒ . Both of the above are probably responsible for the unique perceptual quality of roars compared to trumpets. Out of 51 calls where the age-sex class was clear, 42 ͑ 82.3% ͒ were produced by adult or sub-adult females, six ͑ 11.7% ͒ by juvenile females, and three ͑ 5.8% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 51 cases where the context was clear, seven ͑ 13.7% ͒ were produced during play, four ͑ 7.8% ͒ due to disturbance by humans or vehicles, 28 ͑ 54.9% ͒ in the context of encounters with other non-human species, three ͑ 5.9% ͒ during inter-specific aggressive interactions, and nine ͑ 17.6% ͒ while facing another group or on entering a landscape ͑ Fig. 3 ͒ . Chirps were found to lie in the frequency range of 313– 3370 Hz ͑ Table V ͒ and were produced in a series ͑ Fig. 1, J–K ͒ ranging from 2 to 8 ͑ mean number= 5.2 Ϯ 2.6, n = 25 individuals ͒ in a single bout. The duration of a bout of chirp- ing ranged from 0.68 s to 3.8 s. Spectral envelope analysis revealed up to seven discernible frequency peaks Fig. 2 , with the first two peaks having equal amplitude. Although chirps had seven frequency peaks, the range over which these peaks were distributed was much narrower ͑ Fig. 2 ͒ than in the case of trumpets and roars. Chirps showed significantly lower harmonicity than trumpets and rumbles ͑ Table III, Mann–Whitney U -test, U = 995.5, Z = −4.38, P Ͻ 0.0001, and U = 786.5, Z = −3.8, P Ͻ 0.0001 ͒ . Out of 63 chirp bouts where the age-sex class was clear, 53 ͑ 84% ͒ were produced by adult or sub-adult females and ten ͑ 15.9% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 66 cases where the context was clear, 30 ͑ 45.5% ͒ were due to disturbance by humans or vehicles, 26 ͑ 39.3% ͒ due to ...

Context 2

... other than rumbles were re-sampled at 16 000 Hz whereas rumbles were re- sampled at 600 Hz. LPC analysis was carried out using the autocorrelation method with a time step of 0.005 s and a window length of 0.005 s ͑ 0.05 s for rumbles ͒ . Harmonicity ͑ harmonics-to-noise ratio ͒ was analyzed using the cross- correlation algorithm with a time step of 0.05 s except for chirps, for which 0.005 s was used due to the short duration of the calls. Calls of different age-sex classes adult male, adult female, young male, and young female ͒ were analyzed sepa- rately. The measured features include the call frequency range, duration, harmonicity, as well as mean, minimum, and maximum values of the fundamental frequency ͑ F 0 ͒ ͑ Table I ͒ . In addition, the frequencies and amplitudes of the LPC peaks were measured to examine spectral patterning. The analyzed acoustic features were separated according to the four age-sex classes and compared statistically ͑ for all categories where the sample size was Ն 3 ͒ for each call type using a combination of one-way analysis of variance and pair-wise comparisons of means by unpaired t -tests. The fea- tures compared included mean F 0 , minimum F 0 , maximum A total of 371 calls were recorded from 154 individuals. F 0 , call duration, harmonicity, and frequency and amplitude Of these, 258 calls were analyzed from 109 individuals, con- of the peaks of the LPC spectrum. sisting of 14 males and 95 females ͑ Table II ͒ . The remaining For the rumbles, the start and end frequencies of F 0 , and calls were discarded on account of low signal-to-noise ratio percentage time from the start to the maximum and mini- or overlap with other calls due to simultaneously vocalizing mum frequencies were measured ͑ Table I ͒ . To characterize individuals. Based on structural characteristics, the calls the frequency modulation in finer detail two measures were could be classified into four types, namely, trumpets, chirps, used ͑ Table I ͒ . The fourth harmonic of all calls was used roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, since visual inspection of the spectrograms revealed that the could be distinguished by its unique frequency range ͑ 10– modulation could be measured at high resolution across al- 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the most all recordings whereas the fundamental was sometimes other calls. Chirps were distinguished by their unique tem- contaminated with noise. The higher harmonics may also poral structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of have greater functional relevance in social recognition as ar- much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. gued by McComb et al. ͑ 2003 ͒ based on playback experi- Trumpets and roars differed from chirps in their duration ments on wild African elephants. The first measure was the ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their number of observed peaks in the fourth harmonic of each spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly call, determined visually from the spectrogram with a mini- lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . mum frequency modulation of 4 Hz as a cutoff. The second There were no significant differences between the different age-sex classes in all of the call features that were exam- measure was the frequency change per unit time for the ined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, fourth harmonic, measured by the cumulative change in fre- wherein females had significantly lower mean F 0 and mini- quency divided by the time of the call ͑ window length of 200 mum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, ms ͒ . t = 2.78 ͒ than males ͑ Table III ͒ . To detect the presence of structural subtypes within the rumbles, 13 measured call features ͑ except harmonicity and LPC peaks ͒ were used to generate pair-wise Euclidean distances between calls. The distance matrix was then subjected to cluster analysis using the Unweighted Pair Group Method with Arithmetic mean ͑ UPGMA ͒ ͑ Sneath and Sokal, 1973; Manly, 1986 ͒ . All statistical analyses were performed using STATISTICA ͑ 1999, Statsoft Inc., Tulsa, USA ͒ . A total of 371 calls were recorded from 154 individuals. Of these, 258 calls were analyzed from 109 individuals, con- sisting of 14 males and 95 females ͑ Table II ͒ . The remaining calls were discarded on account of low signal-to-noise ratio or overlap with other calls due to simultaneously vocalizing individuals. Based on structural characteristics, the calls could be classified into four types, namely, trumpets, chirps, roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, could be distinguished by its unique frequency range ͑ 10– 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the other calls. Chirps were distinguished by their unique temporal structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. Trumpets and roars differed from chirps in their duration ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . There were no significant differences between the different age-sex classes in all of the call features that were examined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, wherein females had significantly lower mean F 0 and minimum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, t = 2.78 ͒ than males ͑ Table III ͒ . The different call types were associated with characteristic body postures and behaviors of the vocalizing individual and other members of the herd. These are described in detail in Table IV. Trumpets were loud, conspicuous high-frequency calls. They were in the frequency range of 405–5879 Hz with a mean duration of about 1 s ͑ Table V ͒ . They had a rich harmonic structure with at least seven clearly visible harmonics ͑ Fig. 1, A–C ͒ . Spectral envelope analysis revealed the first frequency peak to be at 706 Hz ͑ Fig. 2 ͒ . The fourth frequency peak ͑ at 3078 Hz ͒ was about 13 dB lower in amplitude than the first frequency. Out of 73 trumpets where the age-sex class was unambiguous, 58 ͑ 79.5% ͒ were produced by adult or sub-adult females, nine ͑ 12.3% ͒ by juvenile females, five ͑ 6.8% ͒ by adult or sub-adult males, and one 1.4% by a juvenile male ͑ Table II ͒ . Out of the 71 trumpets where the context was clear, seven ͑ 9.9% ͒ were in the context of play, 17 ͑ 23.9% ͒ in the context of disturbance by humans or vehicles, 29 ͑ 40.8% ͒ in the context of disturbance by other non-human species, ten ͑ 14% ͒ in the context of inter-specific aggression, and eight ͑ 11.2% ͒ while running out of a waterhole ͑ Fig. 3 ͒ . Spe- cifically, trumpeting was observed while encountering other species such as deer ͑ Axis axis ͒ , gaur ͑ Bos gaurus ͒ , dhole ͑ Cuon alpinus ͒ , bears ͑ Melursus ursinus ͒ , tigers ͑ Panthera tigris ͒ , egrets ͑ Egretta garzetta ͒ , and humans. Roars were noisy, long calls that were in the frequency range of 305–6150 Hz and had a mean duration of about 2 s ͑ Table V, Fig. 1, D–F ͒ . They were in the same frequency range as trumpets and their spectral patterning was also simi- lar with seven frequency peaks Fig. 2 , the first frequency peak being at 656 Hz. The amplitude fell steeply with increasing frequency, the fourth frequency peak being 23 dB below the first ͑ Fig. 2 ͒ . This was in contrast to trumpets, where the fourth frequency peak was about 13 dB below the first ͑ Fig. 2 ͒ . Roars also showed significantly lower harmonicity than trumpets ͑ Table III ͒ . Both of the above are probably responsible for the unique perceptual quality of roars compared to trumpets. Out of 51 calls where the age-sex class was clear, 42 ͑ 82.3% ͒ were produced by adult or sub-adult females, six ͑ 11.7% ͒ by juvenile females, and three ͑ 5.8% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 51 cases where the context was clear, seven ͑ 13.7% ͒ were produced during play, four ͑ 7.8% ͒ due to disturbance by humans or vehicles, 28 ͑ 54.9% ͒ in the context of encounters with other non-human species, three ͑ 5.9% ͒ during inter-specific aggressive interactions, and nine ͑ 17.6% ͒ while facing another group or on entering a landscape ͑ Fig. 3 ͒ . Chirps were found to lie in the frequency range of 313– 3370 Hz ͑ Table V ͒ and were produced in a series ͑ Fig. 1, J–K ͒ ranging from 2 to 8 ͑ mean number= 5.2 Ϯ 2.6, n = 25 individuals ͒ in a single bout. The duration of a bout of chirp- ing ranged from 0.68 s to 3.8 s. Spectral envelope analysis revealed up to seven discernible frequency peaks Fig. 2 , with the first two peaks having equal amplitude. Although chirps had seven frequency peaks, the range over which these peaks were distributed was much narrower ͑ Fig. 2 ͒ than in the case of trumpets and roars. Chirps showed significantly lower harmonicity than trumpets and rumbles ͑ Table III, Mann–Whitney U -test, U = 995.5, Z = −4.38, P Ͻ 0.0001, and U = 786.5, Z = −3.8, P Ͻ 0.0001 ͒ . Out of 63 chirp bouts where the age-sex class was clear, 53 ͑ 84% ͒ were produced by adult or sub-adult females and ten ͑ 15.9% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 66 cases where the context was clear, 30 ͑ 45.5% ͒ were due to disturbance by humans or vehicles, 26 ͑ 39.3% ͒ due to disturbance by other non-human species, four ͑ 6% ͒ in the context of separation of an individual from the herd, ...

Context 3

... grasslands, and within the forests. Surveys were carried out by vehicle and on foot to locate elephant herds during the day. There were a total of 214 encounters with herds during these sur- veys. The herds were classified into three major categories, namely, mixed herds ͑ both females and males in all age classes ͒ , female herds ͑ no adult or sub-adult males ͒ and male herds ͑ adult and sub-adult males only ͒ . Solitary individuals, male or female, were also occasionally encountered. The proportion of adult ͑ Ͼ 15 years ͒ and sub-adult ͑ 5–15 years ͒ males in this population is estimated to be 12.5% ͑ Arivazhagan and Sukumar, 2005 ͒ and approximately 11% from our encounters. Elephants were individually identified based on distinguishing characteristics such as ear-fold, tail characteristics, and pigmentation. The age class of individuals was also determined using methods described elsewhere A narrow-band spectrogram was computed and the minimum and maximum frequencies of the calls were determined using their power spectral densities. To estimate the essential bandwidth, the power spectral density ͑ PSD ͒ of the noisy signal was first computed using a Hamming window ͑ 20 ms ͒ . The PSD of the noise was then estimated using the initial 1 s of the recording. The estimated PSD of the noise was subtracted from that of the noisy signal to yield an esti- mate of the PSD of the signal. In general, the estimate had positive values at all discrete Fourier transform ͑ DFT ͒ bins; any negative values ͑ due to estimation errors ͒ were clamped to zero. From this estimate, the bandpass region containing about 80% of the total energy was considered as the frequency range of the signal. For rumbles, however, calls were first low-pass-filtered with a cutoff of 250 Hz ͑ this value was based on a preliminary inspection of the spectrograms ͒ . The signals thus obtained were re-sampled ͑ by the zero padding approach ͒ to obtain more points along the frequency axis ͑ Oppenheim and Schafer, 1989 ͒ . Spectrograms of these signals ͑ DFT with Hamming window size of 5 ms, 50% overlap ͒ were used for further analysis. Pitch, spectral envelope, and harmonicity analyses were carried out using PRAAT 5.1.07 ͑ www.praat.org, Paul Boersma and David Weenink ͒ . Pitch analysis was carried out using a pitch floor of 10 Hz for rumbles and 100 Hz for most other calls. Spectral envelope analysis was carried out using linear predictive coding ͑ LPC ͒ . Calls other than rumbles were re-sampled at 16 000 Hz whereas rumbles were re- sampled at 600 Hz. LPC analysis was carried out using the autocorrelation method with a time step of 0.005 s and a window length of 0.005 s ͑ 0.05 s for rumbles ͒ . Harmonicity ͑ harmonics-to-noise ratio ͒ was analyzed using the cross- correlation algorithm with a time step of 0.05 s except for chirps, for which 0.005 s was used due to the short duration of the calls. Calls of different age-sex classes adult male, adult female, young male, and young female ͒ were analyzed sepa- rately. The measured features include the call frequency range, duration, harmonicity, as well as mean, minimum, and maximum values of the fundamental frequency ͑ F 0 ͒ ͑ Table I ͒ . In addition, the frequencies and amplitudes of the LPC peaks were measured to examine spectral patterning. The analyzed acoustic features were separated according to the four age-sex classes and compared statistically ͑ for all categories where the sample size was Ն 3 ͒ for each call type using a combination of one-way analysis of variance and pair-wise comparisons of means by unpaired t -tests. The fea- tures compared included mean F 0 , minimum F 0 , maximum A total of 371 calls were recorded from 154 individuals. F 0 , call duration, harmonicity, and frequency and amplitude Of these, 258 calls were analyzed from 109 individuals, con- of the peaks of the LPC spectrum. sisting of 14 males and 95 females ͑ Table II ͒ . The remaining For the rumbles, the start and end frequencies of F 0 , and calls were discarded on account of low signal-to-noise ratio percentage time from the start to the maximum and mini- or overlap with other calls due to simultaneously vocalizing mum frequencies were measured ͑ Table I ͒ . To characterize individuals. Based on structural characteristics, the calls the frequency modulation in finer detail two measures were could be classified into four types, namely, trumpets, chirps, used ͑ Table I ͒ . The fourth harmonic of all calls was used roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, since visual inspection of the spectrograms revealed that the could be distinguished by its unique frequency range ͑ 10– modulation could be measured at high resolution across al- 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the most all recordings whereas the fundamental was sometimes other calls. Chirps were distinguished by their unique tem- contaminated with noise. The higher harmonics may also poral structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of have greater functional relevance in social recognition as ar- much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. gued by McComb et al. ͑ 2003 ͒ based on playback experi- Trumpets and roars differed from chirps in their duration ments on wild African elephants. The first measure was the ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their number of observed peaks in the fourth harmonic of each spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly call, determined visually from the spectrogram with a mini- lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . mum frequency modulation of 4 Hz as a cutoff. The second There were no significant differences between the different age-sex classes in all of the call features that were exam- measure was the frequency change per unit time for the ined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, fourth harmonic, measured by the cumulative change in fre- wherein females had significantly lower mean F 0 and mini- quency divided by the time of the call ͑ window length of 200 mum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, ms ͒ . t = 2.78 ͒ than males ͑ Table III ͒ . To detect the presence of structural subtypes within the rumbles, 13 measured call features ͑ except harmonicity and LPC peaks ͒ were used to generate pair-wise Euclidean distances between calls. The distance matrix was then subjected to cluster analysis using the Unweighted Pair Group Method with Arithmetic mean ͑ UPGMA ͒ ͑ Sneath and Sokal, 1973; Manly, 1986 ͒ . All statistical analyses were performed using STATISTICA ͑ 1999, Statsoft Inc., Tulsa, USA ͒ . A total of 371 calls were recorded from 154 individuals. Of these, 258 calls were analyzed from 109 individuals, con- sisting of 14 males and 95 females ͑ Table II ͒ . The remaining calls were discarded on account of low signal-to-noise ratio or overlap with other calls due to simultaneously vocalizing individuals. Based on structural characteristics, the calls could be classified into four types, namely, trumpets, chirps, roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, could be distinguished by its unique frequency range ͑ 10– 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the other calls. Chirps were distinguished by their unique temporal structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. Trumpets and roars differed from chirps in their duration ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . There were no significant differences between the different age-sex classes in all of the call features that were examined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, wherein females had significantly lower mean F 0 and minimum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, t = 2.78 ͒ than males ͑ Table III ͒ . The different call types were associated with characteristic body postures and behaviors of the vocalizing individual and other members of the herd. These are described in detail in Table IV. Trumpets were loud, conspicuous high-frequency calls. They were in the frequency range of 405–5879 Hz with a mean duration of about 1 s ͑ Table V ͒ . They had a rich harmonic structure with at least seven clearly visible harmonics ͑ Fig. 1, A–C ͒ . Spectral envelope analysis revealed the first frequency peak to be at 706 Hz ͑ Fig. 2 ͒ . The fourth frequency peak ͑ at 3078 Hz ͒ was about 13 dB lower in amplitude than the first frequency. Out of 73 trumpets where the age-sex class was unambiguous, 58 ͑ 79.5% ͒ were produced by adult or sub-adult females, nine ͑ 12.3% ͒ by juvenile females, five ͑ 6.8% ͒ by adult or sub-adult males, and one 1.4% by a juvenile male ͑ Table II ͒ . Out of the 71 trumpets where the context was clear, seven ͑ 9.9% ͒ were in the context of play, 17 ͑ 23.9% ͒ in the context of disturbance by humans or vehicles, 29 ͑ 40.8% ͒ in the context of disturbance by other non-human species, ten ͑ 14% ͒ in the context of inter-specific aggression, and eight ͑ 11.2% ͒ while running out of a waterhole ͑ Fig. 3 ͒ . Spe- cifically, trumpeting was observed while encountering other species such as deer ͑ Axis axis ͒ , gaur ͑ Bos gaurus ͒ , dhole ͑ Cuon alpinus ͒ , bears ͑ Melursus ursinus ͒ , tigers ͑ Panthera tigris ͒ , egrets ͑ Egretta garzetta ͒ , and humans. ...

Context 4

... remaining For the rumbles, the start and end frequencies of F 0 , and calls were discarded on account of low signal-to-noise ratio percentage time from the start to the maximum and mini- or overlap with other calls due to simultaneously vocalizing mum frequencies were measured ͑ Table I ͒ . To characterize individuals. Based on structural characteristics, the calls the frequency modulation in finer detail two measures were could be classified into four types, namely, trumpets, chirps, used ͑ Table I ͒ . The fourth harmonic of all calls was used roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, since visual inspection of the spectrograms revealed that the could be distinguished by its unique frequency range ͑ 10– modulation could be measured at high resolution across al- 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the most all recordings whereas the fundamental was sometimes other calls. Chirps were distinguished by their unique tem- contaminated with noise. The higher harmonics may also poral structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of have greater functional relevance in social recognition as ar- much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. gued by McComb et al. ͑ 2003 ͒ based on playback experi- Trumpets and roars differed from chirps in their duration ments on wild African elephants. The first measure was the ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their number of observed peaks in the fourth harmonic of each spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly call, determined visually from the spectrogram with a mini- lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . mum frequency modulation of 4 Hz as a cutoff. The second There were no significant differences between the different age-sex classes in all of the call features that were exam- measure was the frequency change per unit time for the ined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, fourth harmonic, measured by the cumulative change in fre- wherein females had significantly lower mean F 0 and mini- quency divided by the time of the call ͑ window length of 200 mum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, ms ͒ . t = 2.78 ͒ than males ͑ Table III ͒ . To detect the presence of structural subtypes within the rumbles, 13 measured call features ͑ except harmonicity and LPC peaks ͒ were used to generate pair-wise Euclidean distances between calls. The distance matrix was then subjected to cluster analysis using the Unweighted Pair Group Method with Arithmetic mean ͑ UPGMA ͒ ͑ Sneath and Sokal, 1973; Manly, 1986 ͒ . All statistical analyses were performed using STATISTICA ͑ 1999, Statsoft Inc., Tulsa, USA ͒ . A total of 371 calls were recorded from 154 individuals. Of these, 258 calls were analyzed from 109 individuals, con- sisting of 14 males and 95 females ͑ Table II ͒ . The remaining calls were discarded on account of low signal-to-noise ratio or overlap with other calls due to simultaneously vocalizing individuals. Based on structural characteristics, the calls could be classified into four types, namely, trumpets, chirps, roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, could be distinguished by its unique frequency range ͑ 10– 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the other calls. Chirps were distinguished by their unique temporal structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. Trumpets and roars differed from chirps in their duration ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . There were no significant differences between the different age-sex classes in all of the call features that were examined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, wherein females had significantly lower mean F 0 and minimum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, t = 2.78 ͒ than males ͑ Table III ͒ . The different call types were associated with characteristic body postures and behaviors of the vocalizing individual and other members of the herd. These are described in detail in Table IV. Trumpets were loud, conspicuous high-frequency calls. They were in the frequency range of 405–5879 Hz with a mean duration of about 1 s ͑ Table V ͒ . They had a rich harmonic structure with at least seven clearly visible harmonics ͑ Fig. 1, A–C ͒ . Spectral envelope analysis revealed the first frequency peak to be at 706 Hz ͑ Fig. 2 ͒ . The fourth frequency peak ͑ at 3078 Hz ͒ was about 13 dB lower in amplitude than the first frequency. Out of 73 trumpets where the age-sex class was unambiguous, 58 ͑ 79.5% ͒ were produced by adult or sub-adult females, nine ͑ 12.3% ͒ by juvenile females, five ͑ 6.8% ͒ by adult or sub-adult males, and one 1.4% by a juvenile male ͑ Table II ͒ . Out of the 71 trumpets where the context was clear, seven ͑ 9.9% ͒ were in the context of play, 17 ͑ 23.9% ͒ in the context of disturbance by humans or vehicles, 29 ͑ 40.8% ͒ in the context of disturbance by other non-human species, ten ͑ 14% ͒ in the context of inter-specific aggression, and eight ͑ 11.2% ͒ while running out of a waterhole ͑ Fig. 3 ͒ . Spe- cifically, trumpeting was observed while encountering other species such as deer ͑ Axis axis ͒ , gaur ͑ Bos gaurus ͒ , dhole ͑ Cuon alpinus ͒ , bears ͑ Melursus ursinus ͒ , tigers ͑ Panthera tigris ͒ , egrets ͑ Egretta garzetta ͒ , and humans. Roars were noisy, long calls that were in the frequency range of 305–6150 Hz and had a mean duration of about 2 s ͑ Table V, Fig. 1, D–F ͒ . They were in the same frequency range as trumpets and their spectral patterning was also simi- lar with seven frequency peaks Fig. 2 , the first frequency peak being at 656 Hz. The amplitude fell steeply with increasing frequency, the fourth frequency peak being 23 dB below the first ͑ Fig. 2 ͒ . This was in contrast to trumpets, where the fourth frequency peak was about 13 dB below the first ͑ Fig. 2 ͒ . Roars also showed significantly lower harmonicity than trumpets ͑ Table III ͒ . Both of the above are probably responsible for the unique perceptual quality of roars compared to trumpets. Out of 51 calls where the age-sex class was clear, 42 ͑ 82.3% ͒ were produced by adult or sub-adult females, six ͑ 11.7% ͒ by juvenile females, and three ͑ 5.8% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 51 cases where the context was clear, seven ͑ 13.7% ͒ were produced during play, four ͑ 7.8% ͒ due to disturbance by humans or vehicles, 28 ͑ 54.9% ͒ in the context of encounters with other non-human species, three ͑ 5.9% ͒ during inter-specific aggressive interactions, and nine ͑ 17.6% ͒ while facing another group or on entering a landscape ͑ Fig. 3 ͒ . Chirps were found to lie in the frequency range of 313– 3370 Hz ͑ Table V ͒ and were produced in a series ͑ Fig. 1, J–K ͒ ranging from 2 to 8 ͑ mean number= 5.2 Ϯ 2.6, n = 25 individuals ͒ in a single bout. The duration of a bout of chirp- ing ranged from 0.68 s to 3.8 s. Spectral envelope analysis revealed up to seven discernible frequency peaks Fig. 2 , with the first two peaks having equal amplitude. Although chirps had seven frequency peaks, the range over which these peaks were distributed was much narrower ͑ Fig. 2 ͒ than in the case of trumpets and roars. Chirps showed significantly lower harmonicity than trumpets and rumbles ͑ Table III, Mann–Whitney U -test, U = 995.5, Z = −4.38, P Ͻ 0.0001, and U = 786.5, Z = −3.8, P Ͻ 0.0001 ͒ . Out of 63 chirp bouts where the age-sex class was clear, 53 ͑ 84% ͒ were produced by adult or sub-adult females and ten ͑ 15.9% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 66 cases where the context was clear, 30 ͑ 45.5% ͒ were due to disturbance by humans or vehicles, 26 ͑ 39.3% ͒ due to disturbance by other non-human species, four ͑ 6% ͒ in the context of separation of an individual from the herd, and six ͑ 9% ͒ during intra-group aggression produced by an individual other than those directly involved in the aggression ͑ Fig. 3 ͒ . Rumbles were the only call type in the repertoire with infrasonic components. Rumbles were found to lie in the frequency range of 10–173 Hz, with a mean duration of 5.2 s ͑ Table V ͒ . They had a distinct harmonic structure ͑ Fig. 1, G–I, Table III ͒ , similar to trumpets ͑ U = 1508, Z = 0.616, and P = 0.54 ͒ . Spectral envelope analysis revealed three peaks, with the first peak at 37 Hz Fig. 2 . The power of the third peak was about 11 dB lower than that of the first frequency ͑ Fig. 2 ͒ . Of the 33 cases where the identity of the rumbling individual was unambiguous, 27 ͑ 81.8% ͒ were produced by adult or sub-adult females, two ͑ 6.1% ͒ by juvenile females, and four ͑ 12.1% ͒ by juvenile males ͑ Table II ͒ . Out of 56 cases where the context of rumbling was clear, 20 ͑ 35.7% ͒ were produced due to disturbance by humans or vehicles, six ͑ 10.7% ͒ due to disturbance by non-human species, three ͑ 5.4% ͒ by matriarchs to assemble the group ͑ “let’s go” rumble ͒ , three ͑ 5.4% ͒ in the context of contacting other herd members ͑ contact calls ͒ , and 24 ͑ 42.9% ͒ during intra- and inter-group interactions at close range ͑ Fig. 3 ͒ . Cluster analysis based on the distance matrix of pair- wise measures of overall similarity between calls did not reveal discrete structural groups, suggesting that the variation in call ...

Context 5

... 2.78 ͒ than males ͑ Table III ͒ . To detect the presence of structural subtypes within the rumbles, 13 measured call features ͑ except harmonicity and LPC peaks ͒ were used to generate pair-wise Euclidean distances between calls. The distance matrix was then subjected to cluster analysis using the Unweighted Pair Group Method with Arithmetic mean ͑ UPGMA ͒ ͑ Sneath and Sokal, 1973; Manly, 1986 ͒ . All statistical analyses were performed using STATISTICA ͑ 1999, Statsoft Inc., Tulsa, USA ͒ . A total of 371 calls were recorded from 154 individuals. Of these, 258 calls were analyzed from 109 individuals, con- sisting of 14 males and 95 females ͑ Table II ͒ . The remaining calls were discarded on account of low signal-to-noise ratio or overlap with other calls due to simultaneously vocalizing individuals. Based on structural characteristics, the calls could be classified into four types, namely, trumpets, chirps, roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, could be distinguished by its unique frequency range ͑ 10– 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the other calls. Chirps were distinguished by their unique temporal structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. Trumpets and roars differed from chirps in their duration ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . There were no significant differences between the different age-sex classes in all of the call features that were examined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, wherein females had significantly lower mean F 0 and minimum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, t = 2.78 ͒ than males ͑ Table III ͒ . The different call types were associated with characteristic body postures and behaviors of the vocalizing individual and other members of the herd. These are described in detail in Table IV. Trumpets were loud, conspicuous high-frequency calls. They were in the frequency range of 405–5879 Hz with a mean duration of about 1 s ͑ Table V ͒ . They had a rich harmonic structure with at least seven clearly visible harmonics ͑ Fig. 1, A–C ͒ . Spectral envelope analysis revealed the first frequency peak to be at 706 Hz ͑ Fig. 2 ͒ . The fourth frequency peak ͑ at 3078 Hz ͒ was about 13 dB lower in amplitude than the first frequency. Out of 73 trumpets where the age-sex class was unambiguous, 58 ͑ 79.5% ͒ were produced by adult or sub-adult females, nine ͑ 12.3% ͒ by juvenile females, five ͑ 6.8% ͒ by adult or sub-adult males, and one 1.4% by a juvenile male ͑ Table II ͒ . Out of the 71 trumpets where the context was clear, seven ͑ 9.9% ͒ were in the context of play, 17 ͑ 23.9% ͒ in the context of disturbance by humans or vehicles, 29 ͑ 40.8% ͒ in the context of disturbance by other non-human species, ten ͑ 14% ͒ in the context of inter-specific aggression, and eight ͑ 11.2% ͒ while running out of a waterhole ͑ Fig. 3 ͒ . Spe- cifically, trumpeting was observed while encountering other species such as deer ͑ Axis axis ͒ , gaur ͑ Bos gaurus ͒ , dhole ͑ Cuon alpinus ͒ , bears ͑ Melursus ursinus ͒ , tigers ͑ Panthera tigris ͒ , egrets ͑ Egretta garzetta ͒ , and humans. Roars were noisy, long calls that were in the frequency range of 305–6150 Hz and had a mean duration of about 2 s ͑ Table V, Fig. 1, D–F ͒ . They were in the same frequency range as trumpets and their spectral patterning was also simi- lar with seven frequency peaks Fig. 2 , the first frequency peak being at 656 Hz. The amplitude fell steeply with increasing frequency, the fourth frequency peak being 23 dB below the first ͑ Fig. 2 ͒ . This was in contrast to trumpets, where the fourth frequency peak was about 13 dB below the first ͑ Fig. 2 ͒ . Roars also showed significantly lower harmonicity than trumpets ͑ Table III ͒ . Both of the above are probably responsible for the unique perceptual quality of roars compared to trumpets. Out of 51 calls where the age-sex class was clear, 42 ͑ 82.3% ͒ were produced by adult or sub-adult females, six ͑ 11.7% ͒ by juvenile females, and three ͑ 5.8% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 51 cases where the context was clear, seven ͑ 13.7% ͒ were produced during play, four ͑ 7.8% ͒ due to disturbance by humans or vehicles, 28 ͑ 54.9% ͒ in the context of encounters with other non-human species, three ͑ 5.9% ͒ during inter-specific aggressive interactions, and nine ͑ 17.6% ͒ while facing another group or on entering a landscape ͑ Fig. 3 ͒ . Chirps were found to lie in the frequency range of 313– 3370 Hz ͑ Table V ͒ and were produced in a series ͑ Fig. 1, J–K ͒ ranging from 2 to 8 ͑ mean number= 5.2 Ϯ 2.6, n = 25 individuals ͒ in a single bout. The duration of a bout of chirp- ing ranged from 0.68 s to 3.8 s. Spectral envelope analysis revealed up to seven discernible frequency peaks Fig. 2 , with the first two peaks having equal amplitude. Although chirps had seven frequency peaks, the range over which these peaks were distributed was much narrower ͑ Fig. 2 ͒ than in the case of trumpets and roars. Chirps showed significantly lower harmonicity than trumpets and rumbles ͑ Table III, Mann–Whitney U -test, U = 995.5, Z = −4.38, P Ͻ 0.0001, and U = 786.5, Z = −3.8, P Ͻ 0.0001 ͒ . Out of 63 chirp bouts where the age-sex class was clear, 53 ͑ 84% ͒ were produced by adult or sub-adult females and ten ͑ 15.9% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 66 cases where the context was clear, 30 ͑ 45.5% ͒ were due to disturbance by humans or vehicles, 26 ͑ 39.3% ͒ due to disturbance by other non-human species, four ͑ 6% ͒ in the context of separation of an individual from the herd, and six ͑ 9% ͒ during intra-group aggression produced by an individual other than those directly involved in the aggression ͑ Fig. 3 ͒ . Rumbles were the only call type in the repertoire with infrasonic components. Rumbles were found to lie in the frequency range of 10–173 Hz, with a mean duration of 5.2 s ͑ Table V ͒ . They had a distinct harmonic structure ͑ Fig. 1, G–I, Table III ͒ , similar to trumpets ͑ U = 1508, Z = 0.616, and P = 0.54 ͒ . Spectral envelope analysis revealed three peaks, with the first peak at 37 Hz Fig. 2 . The power of the third peak was about 11 dB lower than that of the first frequency ͑ Fig. 2 ͒ . Of the 33 cases where the identity of the rumbling individual was unambiguous, 27 ͑ 81.8% ͒ were produced by adult or sub-adult females, two ͑ 6.1% ͒ by juvenile females, and four ͑ 12.1% ͒ by juvenile males ͑ Table II ͒ . Out of 56 cases where the context of rumbling was clear, 20 ͑ 35.7% ͒ were produced due to disturbance by humans or vehicles, six ͑ 10.7% ͒ due to disturbance by non-human species, three ͑ 5.4% ͒ by matriarchs to assemble the group ͑ “let’s go” rumble ͒ , three ͑ 5.4% ͒ in the context of contacting other herd members ͑ contact calls ͒ , and 24 ͑ 42.9% ͒ during intra- and inter-group interactions at close range ͑ Fig. 3 ͒ . Cluster analysis based on the distance matrix of pair- wise measures of overall similarity between calls did not reveal discrete structural groups, suggesting that the variation in call structure is graded. Examination of the spectrograms, however, revealed differences between calls, particularly in the direction and extent of frequency modulation. Some calls showed an overall downward modulation of frequency ͓ Fig. 4 ͑ A ͔͒ , others showed little or no frequency modulation ͓ Fig. 4 ͑ B ͔͒ , and some showed an overall upward modulation in frequency ͓ Fig. 4 ͑ C ͔͒ . Yet another type con- tained extensive frequency modulation within the call ͓ Fig. 4 ͑ D ͔͒ . Preliminary comparisons did not reveal any particular correspondence between these features and either age-sex class or behavioral context, but the sample sizes for many of the groups are too small to permit meaningful conclusions at this stage. This study of the vocalizations of wild Asian elephants classifies them into four mutually exclusive categories based on structural features: trumpets, chirps, roars, and rumbles. Three of the four call types, namely, trumpets, chirps, and roars show extensive overlap in their frequency ranges. They are, however, clearly distinguishable from each other by their temporal and/or spectral structures. Trumpets show high harmonicity relative to chirps and roars, which are noisy. Chirps may be distinguished from roars by their characteristic temporal and spectral structures: short durations and frequency peaks over a narrower range. Roars exhibit low harmonicity and have no specific temporal structure. Rumbles, which constitute the fourth call type, do not overlap with any of the other call types in frequency and exhibit a distinct harmonic structure. Rumbles are also much longer in duration compared with the other call types. Our observations can be compared with previous studies on Asian and African elephants. On the basis of auditory assessments in the field and visual assessments of a few spectrograms, McKay ͑ 1973 ͒ classified the vocalizations of wild Asian elephants in Sri Lanka and zoo elephants into three “basic sounds” with eight “resulting sounds,” depending on their modification by change in amplitude, temporal patterning, and stressing of overtones, as well as non-vocal sounds produced in the trunk. However, the spectral and temporal characteristics were not defined. These “basic sounds” ͑ with “resulting sounds” ͒ were growl ͑ growl, rumble, roar, and “motor- cycle” ͒ , squeak ͑ chirp and trumpet ͒ , and snort ͑ “snort” and “boom” ͒ . It is now clear that the categories of resulting sounds such as rumbles ...

Context 6

... ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. gued by McComb et al. ͑ 2003 ͒ based on playback experi- Trumpets and roars differed from chirps in their duration ments on wild African elephants. The first measure was the ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their number of observed peaks in the fourth harmonic of each spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly call, determined visually from the spectrogram with a mini- lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . mum frequency modulation of 4 Hz as a cutoff. The second There were no significant differences between the different age-sex classes in all of the call features that were exam- measure was the frequency change per unit time for the ined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, fourth harmonic, measured by the cumulative change in fre- wherein females had significantly lower mean F 0 and mini- quency divided by the time of the call ͑ window length of 200 mum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, ms ͒ . t = 2.78 ͒ than males ͑ Table III ͒ . To detect the presence of structural subtypes within the rumbles, 13 measured call features ͑ except harmonicity and LPC peaks ͒ were used to generate pair-wise Euclidean distances between calls. The distance matrix was then subjected to cluster analysis using the Unweighted Pair Group Method with Arithmetic mean ͑ UPGMA ͒ ͑ Sneath and Sokal, 1973; Manly, 1986 ͒ . All statistical analyses were performed using STATISTICA ͑ 1999, Statsoft Inc., Tulsa, USA ͒ . A total of 371 calls were recorded from 154 individuals. Of these, 258 calls were analyzed from 109 individuals, con- sisting of 14 males and 95 females ͑ Table II ͒ . The remaining calls were discarded on account of low signal-to-noise ratio or overlap with other calls due to simultaneously vocalizing individuals. Based on structural characteristics, the calls could be classified into four types, namely, trumpets, chirps, roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, could be distinguished by its unique frequency range ͑ 10– 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the other calls. Chirps were distinguished by their unique temporal structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. Trumpets and roars differed from chirps in their duration ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . There were no significant differences between the different age-sex classes in all of the call features that were examined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, wherein females had significantly lower mean F 0 and minimum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, t = 2.78 ͒ than males ͑ Table III ͒ . The different call types were associated with characteristic body postures and behaviors of the vocalizing individual and other members of the herd. These are described in detail in Table IV. Trumpets were loud, conspicuous high-frequency calls. They were in the frequency range of 405–5879 Hz with a mean duration of about 1 s ͑ Table V ͒ . They had a rich harmonic structure with at least seven clearly visible harmonics ͑ Fig. 1, A–C ͒ . Spectral envelope analysis revealed the first frequency peak to be at 706 Hz ͑ Fig. 2 ͒ . The fourth frequency peak ͑ at 3078 Hz ͒ was about 13 dB lower in amplitude than the first frequency. Out of 73 trumpets where the age-sex class was unambiguous, 58 ͑ 79.5% ͒ were produced by adult or sub-adult females, nine ͑ 12.3% ͒ by juvenile females, five ͑ 6.8% ͒ by adult or sub-adult males, and one 1.4% by a juvenile male ͑ Table II ͒ . Out of the 71 trumpets where the context was clear, seven ͑ 9.9% ͒ were in the context of play, 17 ͑ 23.9% ͒ in the context of disturbance by humans or vehicles, 29 ͑ 40.8% ͒ in the context of disturbance by other non-human species, ten ͑ 14% ͒ in the context of inter-specific aggression, and eight ͑ 11.2% ͒ while running out of a waterhole ͑ Fig. 3 ͒ . Spe- cifically, trumpeting was observed while encountering other species such as deer ͑ Axis axis ͒ , gaur ͑ Bos gaurus ͒ , dhole ͑ Cuon alpinus ͒ , bears ͑ Melursus ursinus ͒ , tigers ͑ Panthera tigris ͒ , egrets ͑ Egretta garzetta ͒ , and humans. Roars were noisy, long calls that were in the frequency range of 305–6150 Hz and had a mean duration of about 2 s ͑ Table V, Fig. 1, D–F ͒ . They were in the same frequency range as trumpets and their spectral patterning was also simi- lar with seven frequency peaks Fig. 2 , the first frequency peak being at 656 Hz. The amplitude fell steeply with increasing frequency, the fourth frequency peak being 23 dB below the first ͑ Fig. 2 ͒ . This was in contrast to trumpets, where the fourth frequency peak was about 13 dB below the first ͑ Fig. 2 ͒ . Roars also showed significantly lower harmonicity than trumpets ͑ Table III ͒ . Both of the above are probably responsible for the unique perceptual quality of roars compared to trumpets. Out of 51 calls where the age-sex class was clear, 42 ͑ 82.3% ͒ were produced by adult or sub-adult females, six ͑ 11.7% ͒ by juvenile females, and three ͑ 5.8% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 51 cases where the context was clear, seven ͑ 13.7% ͒ were produced during play, four ͑ 7.8% ͒ due to disturbance by humans or vehicles, 28 ͑ 54.9% ͒ in the context of encounters with other non-human species, three ͑ 5.9% ͒ during inter-specific aggressive interactions, and nine ͑ 17.6% ͒ while facing another group or on entering a landscape ͑ Fig. 3 ͒ . Chirps were found to lie in the frequency range of 313– 3370 Hz ͑ Table V ͒ and were produced in a series ͑ Fig. 1, J–K ͒ ranging from 2 to 8 ͑ mean number= 5.2 Ϯ 2.6, n = 25 individuals ͒ in a single bout. The duration of a bout of chirp- ing ranged from 0.68 s to 3.8 s. Spectral envelope analysis revealed up to seven discernible frequency peaks Fig. 2 , with the first two peaks having equal amplitude. Although chirps had seven frequency peaks, the range over which these peaks were distributed was much narrower ͑ Fig. 2 ͒ than in the case of trumpets and roars. Chirps showed significantly lower harmonicity than trumpets and rumbles ͑ Table III, Mann–Whitney U -test, U = 995.5, Z = −4.38, P Ͻ 0.0001, and U = 786.5, Z = −3.8, P Ͻ 0.0001 ͒ . Out of 63 chirp bouts where the age-sex class was clear, 53 ͑ 84% ͒ were produced by adult or sub-adult females and ten ͑ 15.9% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 66 cases where the context was clear, 30 ͑ 45.5% ͒ were due to disturbance by humans or vehicles, 26 ͑ 39.3% ͒ due to disturbance by other non-human species, four ͑ 6% ͒ in the context of separation of an individual from the herd, and six ͑ 9% ͒ during intra-group aggression produced by an individual other than those directly involved in the aggression ͑ Fig. 3 ͒ . Rumbles were the only call type in the repertoire with infrasonic components. Rumbles were found to lie in the frequency range of 10–173 Hz, with a mean duration of 5.2 s ͑ Table V ͒ . They had a distinct harmonic structure ͑ Fig. 1, G–I, Table III ͒ , similar to trumpets ͑ U = 1508, Z = 0.616, and P = 0.54 ͒ . Spectral envelope analysis revealed three peaks, with the first peak at 37 Hz Fig. 2 . The power of the third peak was about 11 dB lower than that of the first frequency ͑ Fig. 2 ͒ . Of the 33 cases where the identity of the rumbling individual was unambiguous, 27 ͑ 81.8% ͒ were produced by adult or sub-adult females, two ͑ 6.1% ͒ by juvenile females, and four ͑ 12.1% ͒ by juvenile males ͑ Table II ͒ . Out of 56 cases where the context of rumbling was clear, 20 ͑ 35.7% ͒ were produced due to disturbance by humans or vehicles, six ͑ 10.7% ͒ due to disturbance by non-human species, three ͑ 5.4% ͒ by matriarchs to assemble the group ͑ “let’s go” rumble ͒ , three ͑ 5.4% ͒ in the context of contacting other herd members ͑ contact calls ͒ , and 24 ͑ 42.9% ͒ during intra- and inter-group interactions at close range ͑ Fig. 3 ͒ . Cluster analysis based on the distance matrix of pair- wise measures of overall similarity between calls did not reveal discrete structural groups, suggesting that the variation in call structure is graded. Examination of the spectrograms, however, revealed differences between calls, particularly in the direction and extent of frequency modulation. Some calls showed an overall downward modulation of frequency ͓ Fig. 4 ͑ A ͔͒ , others showed little or no frequency modulation ͓ Fig. 4 ͑ B ͔͒ , and some showed an overall upward modulation in frequency ͓ Fig. 4 ͑ C ͔͒ . Yet another type con- tained extensive frequency modulation within the call ͓ Fig. 4 ͑ D ͔͒ . Preliminary comparisons did not reveal any particular correspondence between these features and either age-sex class or behavioral context, but the sample sizes for many of the groups are too small to permit meaningful conclusions at this stage. This study of the vocalizations of wild Asian elephants classifies them into four mutually exclusive categories based on structural features: trumpets, chirps, roars, and rumbles. Three of the four call types, namely, trumpets, chirps, and roars show extensive overlap in their frequency ranges. They are, however, clearly distinguishable from each other by their temporal and/or spectral structures. Trumpets show high harmonicity relative to chirps and ...

Context 7

... and pigmentation. The age class of individuals was also determined using methods described elsewhere Fieldwork was conducted mainly in the dry months ͑ February to May ͒ of 2006 and 2007, in and around waterholes, salt licks, open areas such as swamps or grasslands, and within the forests. Surveys were carried out by vehicle and on foot to locate elephant herds during the day. There were a total of 214 encounters with herds during these sur- veys. The herds were classified into three major categories, namely, mixed herds ͑ both females and males in all age classes ͒ , female herds ͑ no adult or sub-adult males ͒ and male herds ͑ adult and sub-adult males only ͒ . Solitary individuals, male or female, were also occasionally encountered. The proportion of adult ͑ Ͼ 15 years ͒ and sub-adult ͑ 5–15 years ͒ males in this population is estimated to be 12.5% ͑ Arivazhagan and Sukumar, 2005 ͒ and approximately 11% from our encounters. Elephants were individually identified based on distinguishing characteristics such as ear-fold, tail characteristics, and pigmentation. The age class of individuals was also determined using methods described elsewhere A narrow-band spectrogram was computed and the minimum and maximum frequencies of the calls were determined using their power spectral densities. To estimate the essential bandwidth, the power spectral density ͑ PSD ͒ of the noisy signal was first computed using a Hamming window ͑ 20 ms ͒ . The PSD of the noise was then estimated using the initial 1 s of the recording. The estimated PSD of the noise was subtracted from that of the noisy signal to yield an esti- mate of the PSD of the signal. In general, the estimate had positive values at all discrete Fourier transform ͑ DFT ͒ bins; any negative values ͑ due to estimation errors ͒ were clamped to zero. From this estimate, the bandpass region containing about 80% of the total energy was considered as the frequency range of the signal. For rumbles, however, calls were first low-pass-filtered with a cutoff of 250 Hz ͑ this value was based on a preliminary inspection of the spectrograms ͒ . The signals thus obtained were re-sampled ͑ by the zero padding approach ͒ to obtain more points along the frequency axis ͑ Oppenheim and Schafer, 1989 ͒ . Spectrograms of these signals ͑ DFT with Hamming window size of 5 ms, 50% overlap ͒ were used for further analysis. Pitch, spectral envelope, and harmonicity analyses were carried out using PRAAT 5.1.07 ͑ www.praat.org, Paul Boersma and David Weenink ͒ . Pitch analysis was carried out using a pitch floor of 10 Hz for rumbles and 100 Hz for most other calls. Spectral envelope analysis was carried out using linear predictive coding ͑ LPC ͒ . Calls other than rumbles were re-sampled at 16 000 Hz whereas rumbles were re- sampled at 600 Hz. LPC analysis was carried out using the autocorrelation method with a time step of 0.005 s and a window length of 0.005 s ͑ 0.05 s for rumbles ͒ . Harmonicity ͑ harmonics-to-noise ratio ͒ was analyzed using the cross- correlation algorithm with a time step of 0.05 s except for chirps, for which 0.005 s was used due to the short duration of the calls. Calls of different age-sex classes adult male, adult female, young male, and young female ͒ were analyzed sepa- rately. The measured features include the call frequency range, duration, harmonicity, as well as mean, minimum, and maximum values of the fundamental frequency ͑ F 0 ͒ ͑ Table I ͒ . In addition, the frequencies and amplitudes of the LPC peaks were measured to examine spectral patterning. The analyzed acoustic features were separated according to the four age-sex classes and compared statistically ͑ for all categories where the sample size was Ն 3 ͒ for each call type using a combination of one-way analysis of variance and pair-wise comparisons of means by unpaired t -tests. The fea- tures compared included mean F 0 , minimum F 0 , maximum A total of 371 calls were recorded from 154 individuals. F 0 , call duration, harmonicity, and frequency and amplitude Of these, 258 calls were analyzed from 109 individuals, con- of the peaks of the LPC spectrum. sisting of 14 males and 95 females ͑ Table II ͒ . The remaining For the rumbles, the start and end frequencies of F 0 , and calls were discarded on account of low signal-to-noise ratio percentage time from the start to the maximum and mini- or overlap with other calls due to simultaneously vocalizing mum frequencies were measured ͑ Table I ͒ . To characterize individuals. Based on structural characteristics, the calls the frequency modulation in finer detail two measures were could be classified into four types, namely, trumpets, chirps, used ͑ Table I ͒ . The fourth harmonic of all calls was used roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, since visual inspection of the spectrograms revealed that the could be distinguished by its unique frequency range ͑ 10– modulation could be measured at high resolution across al- 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the most all recordings whereas the fundamental was sometimes other calls. Chirps were distinguished by their unique tem- contaminated with noise. The higher harmonics may also poral structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of have greater functional relevance in social recognition as ar- much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. gued by McComb et al. ͑ 2003 ͒ based on playback experi- Trumpets and roars differed from chirps in their duration ments on wild African elephants. The first measure was the ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their number of observed peaks in the fourth harmonic of each spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly call, determined visually from the spectrogram with a mini- lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . mum frequency modulation of 4 Hz as a cutoff. The second There were no significant differences between the different age-sex classes in all of the call features that were exam- measure was the frequency change per unit time for the ined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, fourth harmonic, measured by the cumulative change in fre- wherein females had significantly lower mean F 0 and mini- quency divided by the time of the call ͑ window length of 200 mum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, ms ͒ . t = 2.78 ͒ than males ͑ Table III ͒ . To detect the presence of structural subtypes within the rumbles, 13 measured call features ͑ except harmonicity and LPC peaks ͒ were used to generate pair-wise Euclidean distances between calls. The distance matrix was then subjected to cluster analysis using the Unweighted Pair Group Method with Arithmetic mean ͑ UPGMA ͒ ͑ Sneath and Sokal, 1973; Manly, 1986 ͒ . All statistical analyses were performed using STATISTICA ͑ 1999, Statsoft Inc., Tulsa, USA ͒ . A total of 371 calls were recorded from 154 individuals. Of these, 258 calls were analyzed from 109 individuals, con- sisting of 14 males and 95 females ͑ Table II ͒ . The remaining calls were discarded on account of low signal-to-noise ratio or overlap with other calls due to simultaneously vocalizing individuals. Based on structural characteristics, the calls could be classified into four types, namely, trumpets, chirps, roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, could be distinguished by its unique frequency range ͑ 10– 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the other calls. Chirps were distinguished by their unique temporal structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. Trumpets and roars differed from chirps in their duration ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . There were no significant differences between the different age-sex classes in all of the call features that were examined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, wherein females had significantly lower mean F 0 and minimum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, t = 2.78 ͒ than males ͑ Table III ͒ . The different call types were associated with characteristic body postures and behaviors of the vocalizing individual and other members of the herd. These are described in detail in Table IV. Trumpets were loud, conspicuous high-frequency calls. They were in the frequency range of 405–5879 Hz with a mean duration of about 1 s ͑ Table V ͒ . They had a rich harmonic structure with at least seven clearly visible harmonics ͑ Fig. 1, A–C ͒ . Spectral envelope analysis revealed the first frequency peak to be at 706 Hz ͑ Fig. 2 ͒ . The fourth frequency peak ͑ at 3078 Hz ͒ was about 13 dB lower in amplitude than the first frequency. Out of 73 trumpets where the age-sex class was unambiguous, 58 ͑ 79.5% ͒ were produced by adult or sub-adult females, nine ͑ 12.3% ͒ by juvenile females, five ͑ 6.8% ͒ by adult or sub-adult males, and one 1.4% by a juvenile male ͑ Table II ͒ . Out of the 71 trumpets where the context was clear, seven ͑ 9.9% ͒ were in the context of play, 17 ͑ 23.9% ͒ in the context of disturbance by humans or vehicles, 29 ͑ 40.8% ͒ in the context of disturbance by other non-human species, ten ͑ 14% ͒ in the context of inter-specific aggression, and eight ͑ 11.2% ͒ while running out of a waterhole ͑ Fig. 3 ...

Context 8

... to zero. from our encounters. Elephants were individually identified The onsets and cessations of calls could be determined from based on distinguishing characteristics such as ear-fold, tail the thresholded envelope. characteristics, and pigmentation. The age class of individuals was also determined using methods described elsewhere Fieldwork was conducted mainly in the dry months ͑ February to May ͒ of 2006 and 2007, in and around waterholes, salt licks, open areas such as swamps or grasslands, and within the forests. Surveys were carried out by vehicle and on foot to locate elephant herds during the day. There were a total of 214 encounters with herds during these sur- veys. The herds were classified into three major categories, namely, mixed herds ͑ both females and males in all age classes ͒ , female herds ͑ no adult or sub-adult males ͒ and male herds ͑ adult and sub-adult males only ͒ . Solitary individuals, male or female, were also occasionally encountered. The proportion of adult ͑ Ͼ 15 years ͒ and sub-adult ͑ 5–15 years ͒ males in this population is estimated to be 12.5% ͑ Arivazhagan and Sukumar, 2005 ͒ and approximately 11% from our encounters. Elephants were individually identified based on distinguishing characteristics such as ear-fold, tail characteristics, and pigmentation. The age class of individuals was also determined using methods described elsewhere A narrow-band spectrogram was computed and the minimum and maximum frequencies of the calls were determined using their power spectral densities. To estimate the essential bandwidth, the power spectral density ͑ PSD ͒ of the noisy signal was first computed using a Hamming window ͑ 20 ms ͒ . The PSD of the noise was then estimated using the initial 1 s of the recording. The estimated PSD of the noise was subtracted from that of the noisy signal to yield an esti- mate of the PSD of the signal. In general, the estimate had positive values at all discrete Fourier transform ͑ DFT ͒ bins; any negative values ͑ due to estimation errors ͒ were clamped to zero. From this estimate, the bandpass region containing about 80% of the total energy was considered as the frequency range of the signal. For rumbles, however, calls were first low-pass-filtered with a cutoff of 250 Hz ͑ this value was based on a preliminary inspection of the spectrograms ͒ . The signals thus obtained were re-sampled ͑ by the zero padding approach ͒ to obtain more points along the frequency axis ͑ Oppenheim and Schafer, 1989 ͒ . Spectrograms of these signals ͑ DFT with Hamming window size of 5 ms, 50% overlap ͒ were used for further analysis. Pitch, spectral envelope, and harmonicity analyses were carried out using PRAAT 5.1.07 ͑ www.praat.org, Paul Boersma and David Weenink ͒ . Pitch analysis was carried out using a pitch floor of 10 Hz for rumbles and 100 Hz for most other calls. Spectral envelope analysis was carried out using linear predictive coding ͑ LPC ͒ . Calls other than rumbles were re-sampled at 16 000 Hz whereas rumbles were re- sampled at 600 Hz. LPC analysis was carried out using the autocorrelation method with a time step of 0.005 s and a window length of 0.005 s ͑ 0.05 s for rumbles ͒ . Harmonicity ͑ harmonics-to-noise ratio ͒ was analyzed using the cross- correlation algorithm with a time step of 0.05 s except for chirps, for which 0.005 s was used due to the short duration of the calls. Calls of different age-sex classes adult male, adult female, young male, and young female ͒ were analyzed sepa- rately. The measured features include the call frequency range, duration, harmonicity, as well as mean, minimum, and maximum values of the fundamental frequency ͑ F 0 ͒ ͑ Table I ͒ . In addition, the frequencies and amplitudes of the LPC peaks were measured to examine spectral patterning. The analyzed acoustic features were separated according to the four age-sex classes and compared statistically ͑ for all categories where the sample size was Ն 3 ͒ for each call type using a combination of one-way analysis of variance and pair-wise comparisons of means by unpaired t -tests. The fea- tures compared included mean F 0 , minimum F 0 , maximum A total of 371 calls were recorded from 154 individuals. F 0 , call duration, harmonicity, and frequency and amplitude Of these, 258 calls were analyzed from 109 individuals, con- of the peaks of the LPC spectrum. sisting of 14 males and 95 females ͑ Table II ͒ . The remaining For the rumbles, the start and end frequencies of F 0 , and calls were discarded on account of low signal-to-noise ratio percentage time from the start to the maximum and mini- or overlap with other calls due to simultaneously vocalizing mum frequencies were measured ͑ Table I ͒ . To characterize individuals. Based on structural characteristics, the calls the frequency modulation in finer detail two measures were could be classified into four types, namely, trumpets, chirps, used ͑ Table I ͒ . The fourth harmonic of all calls was used roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, since visual inspection of the spectrograms revealed that the could be distinguished by its unique frequency range ͑ 10– modulation could be measured at high resolution across al- 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the most all recordings whereas the fundamental was sometimes other calls. Chirps were distinguished by their unique tem- contaminated with noise. The higher harmonics may also poral structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of have greater functional relevance in social recognition as ar- much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. gued by McComb et al. ͑ 2003 ͒ based on playback experi- Trumpets and roars differed from chirps in their duration ments on wild African elephants. The first measure was the ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their number of observed peaks in the fourth harmonic of each spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly call, determined visually from the spectrogram with a mini- lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . mum frequency modulation of 4 Hz as a cutoff. The second There were no significant differences between the different age-sex classes in all of the call features that were exam- measure was the frequency change per unit time for the ined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, fourth harmonic, measured by the cumulative change in fre- wherein females had significantly lower mean F 0 and mini- quency divided by the time of the call ͑ window length of 200 mum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, ms ͒ . t = 2.78 ͒ than males ͑ Table III ͒ . To detect the presence of structural subtypes within the rumbles, 13 measured call features ͑ except harmonicity and LPC peaks ͒ were used to generate pair-wise Euclidean distances between calls. The distance matrix was then subjected to cluster analysis using the Unweighted Pair Group Method with Arithmetic mean ͑ UPGMA ͒ ͑ Sneath and Sokal, 1973; Manly, 1986 ͒ . All statistical analyses were performed using STATISTICA ͑ 1999, Statsoft Inc., Tulsa, USA ͒ . A total of 371 calls were recorded from 154 individuals. Of these, 258 calls were analyzed from 109 individuals, con- sisting of 14 males and 95 females ͑ Table II ͒ . The remaining calls were discarded on account of low signal-to-noise ratio or overlap with other calls due to simultaneously vocalizing individuals. Based on structural characteristics, the calls could be classified into four types, namely, trumpets, chirps, roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, could be distinguished by its unique frequency range ͑ 10– 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the other calls. Chirps were distinguished by their unique temporal structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. Trumpets and roars differed from chirps in their duration ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . There were no significant differences between the different age-sex classes in all of the call features that were examined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, wherein females had significantly lower mean F 0 and minimum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, t = 2.78 ͒ than males ͑ Table III ͒ . The different call types were associated with characteristic body postures and behaviors of the vocalizing individual and other members of the herd. These are described in detail in Table IV. Trumpets were loud, conspicuous high-frequency calls. They were in the frequency range of 405–5879 Hz with a mean duration of about 1 s ͑ Table V ͒ . They had a rich harmonic structure with at least seven clearly visible harmonics ͑ Fig. 1, A–C ͒ . Spectral envelope analysis revealed the first frequency peak to be at 706 Hz ͑ Fig. 2 ͒ . The fourth frequency peak ͑ at 3078 Hz ͒ was about 13 dB lower in amplitude than the first frequency. Out of 73 trumpets where the age-sex class was unambiguous, 58 ͑ 79.5% ͒ were produced by adult or sub-adult females, nine ͑ 12.3% ͒ by juvenile females, five ͑ 6.8% ͒ by adult or sub-adult males, and one 1.4% by a juvenile male ͑ Table II ͒ . Out of the 71 trumpets where the context was clear, seven ͑ 9.9% ͒ were in the context of play, 17 ͑ 23.9% ͒ in the ...

Context 9

... roars shared the same frequency range, as revealed by their spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . There were no significant differences between the different age-sex classes in all of the call features that were examined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, wherein females had significantly lower mean F 0 and minimum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, t = 2.78 ͒ than males ͑ Table III ͒ . The different call types were associated with characteristic body postures and behaviors of the vocalizing individual and other members of the herd. These are described in detail in Table IV. Trumpets were loud, conspicuous high-frequency calls. They were in the frequency range of 405–5879 Hz with a mean duration of about 1 s ͑ Table V ͒ . They had a rich harmonic structure with at least seven clearly visible harmonics ͑ Fig. 1, A–C ͒ . Spectral envelope analysis revealed the first frequency peak to be at 706 Hz ͑ Fig. 2 ͒ . The fourth frequency peak ͑ at 3078 Hz ͒ was about 13 dB lower in amplitude than the first frequency. Out of 73 trumpets where the age-sex class was unambiguous, 58 ͑ 79.5% ͒ were produced by adult or sub-adult females, nine ͑ 12.3% ͒ by juvenile females, five ͑ 6.8% ͒ by adult or sub-adult males, and one 1.4% by a juvenile male ͑ Table II ͒ . Out of the 71 trumpets where the context was clear, seven ͑ 9.9% ͒ were in the context of play, 17 ͑ 23.9% ͒ in the context of disturbance by humans or vehicles, 29 ͑ 40.8% ͒ in the context of disturbance by other non-human species, ten ͑ 14% ͒ in the context of inter-specific aggression, and eight ͑ 11.2% ͒ while running out of a waterhole ͑ Fig. 3 ͒ . Spe- cifically, trumpeting was observed while encountering other species such as deer ͑ Axis axis ͒ , gaur ͑ Bos gaurus ͒ , dhole ͑ Cuon alpinus ͒ , bears ͑ Melursus ursinus ͒ , tigers ͑ Panthera tigris ͒ , egrets ͑ Egretta garzetta ͒ , and humans. Roars were noisy, long calls that were in the frequency range of 305–6150 Hz and had a mean duration of about 2 s ͑ Table V, Fig. 1, D–F ͒ . They were in the same frequency range as trumpets and their spectral patterning was also simi- lar with seven frequency peaks Fig. 2 , the first frequency peak being at 656 Hz. The amplitude fell steeply with increasing frequency, the fourth frequency peak being 23 dB below the first ͑ Fig. 2 ͒ . This was in contrast to trumpets, where the fourth frequency peak was about 13 dB below the first ͑ Fig. 2 ͒ . Roars also showed significantly lower harmonicity than trumpets ͑ Table III ͒ . Both of the above are probably responsible for the unique perceptual quality of roars compared to trumpets. Out of 51 calls where the age-sex class was clear, 42 ͑ 82.3% ͒ were produced by adult or sub-adult females, six ͑ 11.7% ͒ by juvenile females, and three ͑ 5.8% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 51 cases where the context was clear, seven ͑ 13.7% ͒ were produced during play, four ͑ 7.8% ͒ due to disturbance by humans or vehicles, 28 ͑ 54.9% ͒ in the context of encounters with other non-human species, three ͑ 5.9% ͒ during inter-specific aggressive interactions, and nine ͑ 17.6% ͒ while facing another group or on entering a landscape ͑ Fig. 3 ͒ . Chirps were found to lie in the frequency range of 313– 3370 Hz ͑ Table V ͒ and were produced in a series ͑ Fig. 1, J–K ͒ ranging from 2 to 8 ͑ mean number= 5.2 Ϯ 2.6, n = 25 individuals ͒ in a single bout. The duration of a bout of chirp- ing ranged from 0.68 s to 3.8 s. Spectral envelope analysis revealed up to seven discernible frequency peaks Fig. 2 , with the first two peaks having equal amplitude. Although chirps had seven frequency peaks, the range over which these peaks were distributed was much narrower ͑ Fig. 2 ͒ than in the case of trumpets and roars. Chirps showed significantly lower harmonicity than trumpets and rumbles ͑ Table III, Mann–Whitney U -test, U = 995.5, Z = −4.38, P Ͻ 0.0001, and U = 786.5, Z = −3.8, P Ͻ 0.0001 ͒ . Out of 63 chirp bouts where the age-sex class was clear, 53 ͑ 84% ͒ were produced by adult or sub-adult females and ten ͑ 15.9% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 66 cases where the context was clear, 30 ͑ 45.5% ͒ were due to disturbance by humans or vehicles, 26 ͑ 39.3% ͒ due to disturbance by other non-human species, four ͑ 6% ͒ in the context of separation of an individual from the herd, and six ͑ 9% ͒ during intra-group aggression produced by an individual other than those directly involved in the aggression ͑ Fig. 3 ͒ . Rumbles were the only call type in the repertoire with infrasonic components. Rumbles were found to lie in the frequency range of 10–173 Hz, with a mean duration of 5.2 s ͑ Table V ͒ . They had a distinct harmonic structure ͑ Fig. 1, G–I, Table III ͒ , similar to trumpets ͑ U = 1508, Z = 0.616, and P = 0.54 ͒ . Spectral envelope analysis revealed three peaks, with the first peak at 37 Hz Fig. 2 . The power of the third peak was about 11 dB lower than that of the first frequency ͑ Fig. 2 ͒ . Of the 33 cases where the identity of the rumbling individual was unambiguous, 27 ͑ 81.8% ͒ were produced by adult or sub-adult females, two ͑ 6.1% ͒ by juvenile females, and four ͑ 12.1% ͒ by juvenile males ͑ Table II ͒ . Out of 56 cases where the context of rumbling was clear, 20 ͑ 35.7% ͒ were produced due to disturbance by humans or vehicles, six ͑ 10.7% ͒ due to disturbance by non-human species, three ͑ 5.4% ͒ by matriarchs to assemble the group ͑ “let’s go” rumble ͒ , three ͑ 5.4% ͒ in the context of contacting other herd members ͑ contact calls ͒ , and 24 ͑ 42.9% ͒ during intra- and inter-group interactions at close range ͑ Fig. 3 ͒ . Cluster analysis based on the distance matrix of pair- wise measures of overall similarity between calls did not reveal discrete structural groups, suggesting that the variation in call structure is graded. Examination of the spectrograms, however, revealed differences between calls, particularly in the direction and extent of frequency modulation. Some calls showed an overall downward modulation of frequency ͓ Fig. 4 ͑ A ͔͒ , others showed little or no frequency modulation ͓ Fig. 4 ͑ B ͔͒ , and some showed an overall upward modulation in frequency ͓ Fig. 4 ͑ C ͔͒ . Yet another type con- tained extensive frequency modulation within the call ͓ Fig. 4 ͑ D ͔͒ . Preliminary comparisons did not reveal any particular correspondence between these features and either age-sex class or behavioral context, but the sample sizes for many of the groups are too small to permit meaningful conclusions at this stage. This study of the vocalizations of wild Asian elephants classifies them into four mutually exclusive categories based on structural features: trumpets, chirps, roars, and rumbles. Three of the four call types, namely, trumpets, chirps, and roars show extensive overlap in their frequency ranges. They are, however, clearly distinguishable from each other by their temporal and/or spectral structures. Trumpets show high harmonicity relative to chirps and roars, which are noisy. Chirps may be distinguished from roars by their characteristic temporal and spectral structures: short durations and frequency peaks over a narrower range. Roars exhibit low harmonicity and have no specific temporal structure. Rumbles, which constitute the fourth call type, do not overlap with any of the other call types in frequency and exhibit a distinct harmonic structure. Rumbles are also much longer in duration compared with the other call types. Our observations can be compared with previous studies on Asian and African elephants. On the basis of auditory assessments in the field and visual assessments of a few spectrograms, McKay ͑ 1973 ͒ classified the vocalizations of wild Asian elephants in Sri Lanka and zoo elephants into three “basic sounds” with eight “resulting sounds,” depending on their modification by change in amplitude, temporal patterning, and stressing of overtones, as well as non-vocal sounds produced in the trunk. However, the spectral and temporal characteristics were not defined. These “basic sounds” ͑ with “resulting sounds” ͒ were growl ͑ growl, rumble, roar, and “motor- cycle” ͒ , squeak ͑ chirp and trumpet ͒ , and snort ͑ “snort” and “boom” ͒ . It is now clear that the categories of resulting sounds such as rumbles and motorcycle with infrasonic components are structurally different from roars, which are calls with low harmonicity and no infrasonic frequencies. Simi- larly, the chirp and the trumpet are sufficiently different in their spectrograms not to be placed together under the basic sound squeak. Non-vocal sounds such as snort and boom are not considered here since our study was confined to vocalizations. Several observers including Sanderson ͑ 1878 ͒ , Krishnan ͑ 1972 ͒ and McKay ͑ 1973 ͒ also described calls of Asian elephants that clearly indicate low frequency sounds. A study on captive female Asian elephants by Payne et al. ͑ 1986 ͒ later recorded infrasound with fundamental frequencies of 14–24 Hz and 10–15 s duration. There have been a number of attempts at classifying African elephant vocalizations, with most of the studies fo- cusing on infrasound. One of the earliest studies on the African species ͑ Berg, 1983 ͒ described the characteristics of the vocalizations of a group of captive elephants based on visual inspection of spectrograms and divided them into ten call types. Our recordings of Asian elephant vocalizations show correspondence with three of these ten types, namely, trumpets, roars, and barks ͓ which we refer to as chirps in accor- dance with McKay ͑ 1973 ͔͒ . The trumpets and roars recorded by Berg ͑ 1983 ͒ are longer in duration ͑ 2s and 3.8s ͒ but similar in terms of dominant ...

Context 10

... method with a time step of 0.005 s and a window length of 0.005 s ͑ 0.05 s for rumbles ͒ . Harmonicity ͑ harmonics-to-noise ratio ͒ was analyzed using the cross- correlation algorithm with a time step of 0.05 s except for chirps, for which 0.005 s was used due to the short duration of the calls. Calls of different age-sex classes adult male, adult female, young male, and young female ͒ were analyzed sepa- rately. The measured features include the call frequency range, duration, harmonicity, as well as mean, minimum, and maximum values of the fundamental frequency ͑ F 0 ͒ ͑ Table I ͒ . In addition, the frequencies and amplitudes of the LPC peaks were measured to examine spectral patterning. The analyzed acoustic features were separated according to the four age-sex classes and compared statistically ͑ for all categories where the sample size was Ն 3 ͒ for each call type using a combination of one-way analysis of variance and pair-wise comparisons of means by unpaired t -tests. The fea- tures compared included mean F 0 , minimum F 0 , maximum A total of 371 calls were recorded from 154 individuals. F 0 , call duration, harmonicity, and frequency and amplitude Of these, 258 calls were analyzed from 109 individuals, con- of the peaks of the LPC spectrum. sisting of 14 males and 95 females ͑ Table II ͒ . The remaining For the rumbles, the start and end frequencies of F 0 , and calls were discarded on account of low signal-to-noise ratio percentage time from the start to the maximum and mini- or overlap with other calls due to simultaneously vocalizing mum frequencies were measured ͑ Table I ͒ . To characterize individuals. Based on structural characteristics, the calls the frequency modulation in finer detail two measures were could be classified into four types, namely, trumpets, chirps, used ͑ Table I ͒ . The fourth harmonic of all calls was used roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, since visual inspection of the spectrograms revealed that the could be distinguished by its unique frequency range ͑ 10– modulation could be measured at high resolution across al- 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the most all recordings whereas the fundamental was sometimes other calls. Chirps were distinguished by their unique tem- contaminated with noise. The higher harmonics may also poral structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of have greater functional relevance in social recognition as ar- much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. gued by McComb et al. ͑ 2003 ͒ based on playback experi- Trumpets and roars differed from chirps in their duration ments on wild African elephants. The first measure was the ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their number of observed peaks in the fourth harmonic of each spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly call, determined visually from the spectrogram with a mini- lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . mum frequency modulation of 4 Hz as a cutoff. The second There were no significant differences between the different age-sex classes in all of the call features that were exam- measure was the frequency change per unit time for the ined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, fourth harmonic, measured by the cumulative change in fre- wherein females had significantly lower mean F 0 and mini- quency divided by the time of the call ͑ window length of 200 mum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, ms ͒ . t = 2.78 ͒ than males ͑ Table III ͒ . To detect the presence of structural subtypes within the rumbles, 13 measured call features ͑ except harmonicity and LPC peaks ͒ were used to generate pair-wise Euclidean distances between calls. The distance matrix was then subjected to cluster analysis using the Unweighted Pair Group Method with Arithmetic mean ͑ UPGMA ͒ ͑ Sneath and Sokal, 1973; Manly, 1986 ͒ . All statistical analyses were performed using STATISTICA ͑ 1999, Statsoft Inc., Tulsa, USA ͒ . A total of 371 calls were recorded from 154 individuals. Of these, 258 calls were analyzed from 109 individuals, con- sisting of 14 males and 95 females ͑ Table II ͒ . The remaining calls were discarded on account of low signal-to-noise ratio or overlap with other calls due to simultaneously vocalizing individuals. Based on structural characteristics, the calls could be classified into four types, namely, trumpets, chirps, roars, and rumbles ͑ Fig. 1 ͒ . One of the call types, the rumble, could be distinguished by its unique frequency range ͑ 10– 173 Hz, Fig. 1, G–I ͒ , which was much lower than that of the other calls. Chirps were distinguished by their unique temporal structure ͑ Table III, Fig. 1, J–K ͒ : they were typically of much shorter duration ͑ 0.2 s Ϯ 0.1 s ͒ than the other calls. Trumpets and roars differed from chirps in their duration ͑ mean= 1 s and 2 s, Table III ͒ . Although both trumpets and roars shared the same frequency range, as revealed by their spectral peaks ͑ Fig. 2 ͒ , the decrease in power with increasing frequency was steeper in roars. Roars also had significantly lower harmonicity than trumpets ͑ Table III, Mann–Whitney U test, U = 313, Z = −7.8, and P Ͻ 0.0001 ͒ . There were no significant differences between the different age-sex classes in all of the call features that were examined in roars, chirps, and rumbles. The only significant differences were between adult male and female trumpets, wherein females had significantly lower mean F 0 and minimum F 0 ͑ unpaired t -tests, P = 0.015, t = 2.45 and P = 0.043, t = 2.78 ͒ than males ͑ Table III ͒ . The different call types were associated with characteristic body postures and behaviors of the vocalizing individual and other members of the herd. These are described in detail in Table IV. Trumpets were loud, conspicuous high-frequency calls. They were in the frequency range of 405–5879 Hz with a mean duration of about 1 s ͑ Table V ͒ . They had a rich harmonic structure with at least seven clearly visible harmonics ͑ Fig. 1, A–C ͒ . Spectral envelope analysis revealed the first frequency peak to be at 706 Hz ͑ Fig. 2 ͒ . The fourth frequency peak ͑ at 3078 Hz ͒ was about 13 dB lower in amplitude than the first frequency. Out of 73 trumpets where the age-sex class was unambiguous, 58 ͑ 79.5% ͒ were produced by adult or sub-adult females, nine ͑ 12.3% ͒ by juvenile females, five ͑ 6.8% ͒ by adult or sub-adult males, and one 1.4% by a juvenile male ͑ Table II ͒ . Out of the 71 trumpets where the context was clear, seven ͑ 9.9% ͒ were in the context of play, 17 ͑ 23.9% ͒ in the context of disturbance by humans or vehicles, 29 ͑ 40.8% ͒ in the context of disturbance by other non-human species, ten ͑ 14% ͒ in the context of inter-specific aggression, and eight ͑ 11.2% ͒ while running out of a waterhole ͑ Fig. 3 ͒ . Spe- cifically, trumpeting was observed while encountering other species such as deer ͑ Axis axis ͒ , gaur ͑ Bos gaurus ͒ , dhole ͑ Cuon alpinus ͒ , bears ͑ Melursus ursinus ͒ , tigers ͑ Panthera tigris ͒ , egrets ͑ Egretta garzetta ͒ , and humans. Roars were noisy, long calls that were in the frequency range of 305–6150 Hz and had a mean duration of about 2 s ͑ Table V, Fig. 1, D–F ͒ . They were in the same frequency range as trumpets and their spectral patterning was also simi- lar with seven frequency peaks Fig. 2 , the first frequency peak being at 656 Hz. The amplitude fell steeply with increasing frequency, the fourth frequency peak being 23 dB below the first ͑ Fig. 2 ͒ . This was in contrast to trumpets, where the fourth frequency peak was about 13 dB below the first ͑ Fig. 2 ͒ . Roars also showed significantly lower harmonicity than trumpets ͑ Table III ͒ . Both of the above are probably responsible for the unique perceptual quality of roars compared to trumpets. Out of 51 calls where the age-sex class was clear, 42 ͑ 82.3% ͒ were produced by adult or sub-adult females, six ͑ 11.7% ͒ by juvenile females, and three ͑ 5.8% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 51 cases where the context was clear, seven ͑ 13.7% ͒ were produced during play, four ͑ 7.8% ͒ due to disturbance by humans or vehicles, 28 ͑ 54.9% ͒ in the context of encounters with other non-human species, three ͑ 5.9% ͒ during inter-specific aggressive interactions, and nine ͑ 17.6% ͒ while facing another group or on entering a landscape ͑ Fig. 3 ͒ . Chirps were found to lie in the frequency range of 313– 3370 Hz ͑ Table V ͒ and were produced in a series ͑ Fig. 1, J–K ͒ ranging from 2 to 8 ͑ mean number= 5.2 Ϯ 2.6, n = 25 individuals ͒ in a single bout. The duration of a bout of chirp- ing ranged from 0.68 s to 3.8 s. Spectral envelope analysis revealed up to seven discernible frequency peaks Fig. 2 , with the first two peaks having equal amplitude. Although chirps had seven frequency peaks, the range over which these peaks were distributed was much narrower ͑ Fig. 2 ͒ than in the case of trumpets and roars. Chirps showed significantly lower harmonicity than trumpets and rumbles ͑ Table III, Mann–Whitney U -test, U = 995.5, Z = −4.38, P Ͻ 0.0001, and U = 786.5, Z = −3.8, P Ͻ 0.0001 ͒ . Out of 63 chirp bouts where the age-sex class was clear, 53 ͑ 84% ͒ were produced by adult or sub-adult females and ten ͑ 15.9% ͒ by adult or sub-adult males ͑ Table II ͒ . Out of 66 cases where the context was clear, 30 ͑ 45.5% ͒ were due to disturbance by humans or vehicles, 26 ͑ 39.3% ͒ due to disturbance by other non-human species, four ͑ 6% ͒ in the context of separation of an individual from the herd, and six ͑ 9% ͒ during intra-group aggression produced by an individual other than those directly involved in the aggression ͑ Fig. 3 ͒ . ...