Mating Call Recognition in the Green Treefrog (Hyla Cinerea): The Significance of Some Fine-Temporal Properties

Green treefrogs (Hyla cinerea) communicate by means of acoustically complex and fairly stereotyped vocal signals. The male calls at night from a relatively stationary position, and the female locates him and initiates sexual contact. Electronically generated calls which attract gravid females as effectively as a typical, natural mating call share several spectral properties with the natural signal (Gerhardt, 1974a, 1976). Here I present the results of experiments designed to identify pertinent fine-temporal properties of the mating call and to determine how much each property can be varied without jeopardizing the relative attractiveness of the sound. The term ‘fine-temporal’ refers to the detailed structure of the amplitude-time waveform. The period (or lack thereof) with which the waveform repeats and the degree to which the periodicity is reflected in the overall amplitude-time envelope are two fine-temporal properties of particular concern in this study.

The broad aim of this research is to derive a detailed, quantitative set of specifications for sound pattern recognition in the green treefrog. This information will complement and guide research into the underlying neurophysiological mechanisms and will provide some answers to evolutionary questions about the treefrog communication system. Fine-temporal patterns, for example, are not only relevant for species identification but they also serve to differentiate signals within the repertoire of the green treefrog.

Previous studies (Gerhardt, 1974a, 1976) have established that two spectral peaks (one around 1 kHz and the other around 3 kHz) are pertinent for mating-call recognition at a moderate sound pressure level (SPL). For this study I have used three methods to vary systematically the fine-temporal properties of synthetic mating calls while maintaining the spectral structure :

Pulse trains (pulse duration < 10 µs) from a Tektronix 501 pulse generator were filtered with a graphic spectrum equalizer (Briiel and Kjaer 125) and bandpass filter (Krohn-Hite 3202). The third-octave attenuator settings of the equalizer and the cut-off frequencies of the bandpass filter were held constant, and the pulse-repetition rate of the pulse train was varied. Pink noise (General Radio 1309B noise generator) was filtered in the same fashion. Oscillograms and amplitude-frequency spectra of these stimuli and a typical, natural mating call are shown in Fig. 1.

Two phase-locked, summed components (0·9 and 2·7 kHz) were sinusoidally amplitude-modulated at a constant depth (∼ 50%; the voltages of the two carriers relative to the voltages of the modulating sinusoids were held constant). Oscillograms and amplitude-frequency spectra are shown in Fig. 2.

In another series of synthetic calls, three phase-locked, summed components (0·9 +2·7 +3·0 kHz) were sinusoidally amplitude-modulated at a rate of 50/s; the depth of modulation was varied. Two such stimuli were used and are compared with the standard, unmodulated call in Fig. 3.

The beat-frequency of synthetic calls was varied by adjusting the frequencies of phase-locked sinusoids, and comparable stimuli without beats were also generated (Fig. 4).

The waveforms described above were packaged electronically into calls which lasted 0·16 s and repeated every 0·8 second. Each kind of sound was recorded on magnetic tape on one channel of a Nagra stereo recorder and had a fixed timing relationship to another synthetic (or natural) call on the other channel. Oscillograms which illustrate these gross temporal properties are published elsewhere (Gerhardt, 1974a).

A Nagra stereo recorder drove two Nagra DH speaker-amplifiers placed 2 or 4 m apart. Non-linearities in the system were taken into account during the synthesis of experimental stimuli so that the system was flat to within ± 2 dB between 600 and 5000 Hz at a point midway between the speakers on a hard, reflecting surface.

Experiments took place outdoors on a level cement surface (driveway or patio) and indoors on a tile floor. Soundfoam (Scott Pyrell) anechoic wedges were used indoors to reduce sound reflexions in the test area. Sound pressure levels (SPL) were equalized (unless otherwise stated) midway between the speakers with a precision sound level meter (Brüel and Kjaer 2209 or General Radio 1933). All experiments done at an SPL of 60 dB were conducted indoors, where the ambient noise level was less than 50 dB SPL. In 1976 and 1977 a tunable filter (Brüel and Kjaer 3 %-one-third-octave) was used in conjunction with a sound-level meter to verify the frequency-response of the system under actual test conditions.

Gravid female treefrogs, usually found in amplexus, were collected near Savannah, Georgia. Most were refrigerated on the night of capture to inhibit oviposition and were tested within 24−48 h. Each animal was tested at night when the air temperature was between 21 and 26 °C. I released each female midway between the two speakers, each of which alternately emitted a different sound. A responsive female usually moved promptly (within 2 min) to a speaker and touched it. A minimum of eight females was tested in each two-stimulus experiment. Females were usually tested in more than one such experiment, but they were seldom exposed to the same stimuli twice and rarely in succession. Previous exposure to a stimulus failed to bias the choice of a female in a subsequent test using the same stimulus (Gerhardt, unpublished data). Other details of the experimental procedures are published elsewhere (Gerhardt, 1974 a).

Two hundred and eighty-eight female treefrogs (more than 90% of the total tested ; only one stimulus pair, discussed below, accounted for a higher than normal percentage of non-responding females) responded in at least one of the discrimination experiments. The animals were tested over the following six summer seasons (numbers responding in brackets): 1972 [12]; 1973 [13]; 1974 [116]; 1975 [57]; 1976 [57] and 1977 [33]-

The results of experiments using the stimuli shown in Fig. 1 are given in Table 1. The standard synthetic call in this series had a waveform periodicity of 300/s but lacked the pulsatile, initial beginning typical of the natural call. Nevertheless, the synthetic call competed effectively with a natural call. Synthetic calls with a periodicity of 450/s were as attractive as the standard call, but stimuli with periodicities of 50/s, 225/s and 900/s were less attractive. Females also preferred the 300/s standard to an aperiodic call (filtered noise). Nine of 12 females tested in 1972 chose filtered pulses with a repetition rate of 333/s over filtered noise. Combining this result with that presented in Table 1, the total number of females choosing a periodic stimulus (around 300/s) was 22 and the total number choosing filtered noise was seven (P = 0·008, two-tailed binomial test). The aperiodic call and calls with periodicities of 225/s and 900/s were, however, clearly preferred to calls with a periodicity of 100/s. Some of these experiments were flawed by the fact that maximum peak SPL’s rather than root-mean-square (RMS) levels were equalized at the release point. The RMS levels are presented in Table 1 ; the most serious discrepancies involved the 50/s and 100/s calls which were 3−6 dB lower in SPL than calls with periodicities of 225/s and higher. Nevertheless, the discrimination against these sounds was almost certainly not based solely on the discrepancy in RMS levels. Eleven females were given a choice between the 50/s and 100/s calls, but only one female responded to either sound (50/s). Most of these animals made vigorous responses to other synthetic calls in both previous and subsequent tests. Furthermore, females discriminated against the 50/s filtered-pulse calls when the alternative was a sound with a beat-frequency of 60/s (Table 1); in this experiment the RMS levels of the two stimuli were equalized and the maximum peak level of the filtered pulse call exceeded that of the beat-frequency call.

Experiments with synthetic calls with variable rates of sinusoidal amplitudemodulation corroborated the results of experiments with filtered pulses (Table 2), except that the results of the AM300 versus AM225 experiment were not statistically significant. However, the majority (8 of 10) of the females chose the stimulus with the periodicity of 300/s as they did in the experiment with the filtered-pulse calls. In these experiments, the RMS levels of the sounds were equalized ; the maximum peak level of the 50/s 95 % AM call was 2 dB greater than that of the unmodulated standard; this was the greatest discrepancy in maximum peak levels resulting from the equalization of RMS levels.

The relative unattractiveness of sounds having periodicities of 100/s and lower was again demonstrated in experiments in which the beat-frequencies were varied (Table 3). The 300/s standard (0·9 +2·7 +3·0 kHz) was preferred to a call with 60/s beats (0·9+2·84+3·00 kHz); the combination of 0·8+1·1 kHz (300/s) was chosen over the combination of 0·9 + 1·0 kHz (100/s).

The results of experiments in which the depth of modulation of the standard call (0·9 +2·7 +3·0 kHz) was varied are presented in Table 2. Females discriminated against the pulsed signals when the depth of modulation was 50 % and the SPL was at least 75 dB, but not when the depth of modulation was 12%.

Although females preferred the 300/s standard call in the filtered pulse series over the 900/s call (Table 1), there were also significant differences in the spectral structures of the two sounds. The P900 stimulus can also be described as a harmonic series (with a missing second harmonic) with a fundamental frequency of 900 Hz ; thus, there are only two components (2·7 and 3·6 kHz) in the high-frequency band as compared with four components in that region of the spectrum in P300 (2·4, 2·7, 3·0 and 3·3 kHz). I tried to generalize the result by using other kinds of synthetic calls. First, females were given a choice between a sinusoidally modulated (300/s) 900 Hz tone burst and an unmodulated 900 Hz tone burst. They responded exclusively to the modulated sound (Table 2). Secondly, females were given a choice between a two-component call with a waveform periodicity of 300/8 (0·9+3·0 kHz; Fig. 4A) and another two-component call with a periodicity of 900/s (0·9 +2·7 kHz, Fig. 4B). Surprisingly, the animals failed to discriminate between these two sounds (Table 3); the difference in waveform periodicity alone was insufficient to elicit preferential responses. The results of experiments in which the depth of 50/s amplitude-modulation was varied indicated that a minimum change in the overall amplitude-time envelope was importans for differential responses to sounds with different waveform periodicities. I therefore gave females a choice between a call in which the 300/s periodicity was reflected in its amplitude-time envelope, i.e. the standard call with beats (0·9 +2·7 +3·0 kHz) and another call with a periodicity of 300/s but lacking beats (0·9 +3·0 kHz). The preference for the standard call was clearly demonstrated over a 20 dB range of SPL (Table 3). Furthermore, even when the standard call was played back at 75 dB and the level of the 0·9 +3·0 kHz call was increased to 80 dB, females still chose the standard call without exception.

The mean waveform periodicity in the calls of H. cinerea from the eastern United States has been shown to be about 300/s however, this property of the calls varied considerably within and among individuals (Oldham & Gerhardt, 1975; table 1; figure 5). In the experiments reported in this paper female green treefrogs preferred sounds having a waveform periodicity of about 300/s and they strongly discriminated against sounds with periodicities of 100/s and below. They discriminated less strongly against aperiodic calls and calls with periodicities of 225/3 and 900/s. However, an important hypothesis which emerged from these experiments is that periodicity preferences depend on corresponding changes in the amplitude-time envelope which must exceed some minimum value. This idea is prompted by the results of three experiments.

First, females failed to discriminate between the unmodulated standard call and the call in which the depth of modulation (50/s AM) was only 12%; they discriminated when the depth was increased to 50% (Table 2). Secondly, females displayed a strong preference for the standard call with 300/s beats over the call having the same waveform periodicity but lacking beats (Table 3). Thirdly, females failed to discriminate between sounds which had a threefold difference in waveform periodicity but which lacked prominent, corresponding changes in their amplitude-time envelopes (Table 3).

Although these results support the hypothesis, other interpretations are possible because of spectral complications. Changes in the temporal structure of most acoustic signals inevitably involve changes in their spectral structure (frequency and/or phase); indeed, these are equivalent descriptions of any periodic function. When the standard call is amplitude-modulated, for example, energy is introduced as sidebands above and below each component in the unmodulated sound; as the depth of modulation is increased, the magnitude of the sidebands increases. Even though the standard call (0·9 +2·7 +3·0 kHz) and the call without beats (0·9 +3·0 kHz) did not differ significantly in relative amplitudes of their spectral peaks,¹ the second component (2·7 kHz) in the high-frequency region may cause a greater excitation of the auditory neurones tuned to this frequency band (see below) than the single (3-0 kHz) component alone. This seems unlikely since the preference for the call with beats was unchanged when the call without beats was played back at an SPL 5 dB greater. Furthermore, other experiments (Gerhardt, 1974a; unpublished data) indicated that the addition of components to either frequency band or both did not increase the attractiveness of a synthetic call relative to a synthetic call which had at least two components in one band (the two necessary to yield 300/s beats). Nevertheless, these arguments do not eliminate the possibility that the preferential responses of the females in the experiments reported in this paper were based on properties of the sounds other than, or in addition to, the temporal differences. This dilemma is reminiscent of the periodicity-pitch controversy in human psychophysics: the same ambiguity regarding the basis (time domain versus frequency domain) of discrimination often confounds the interpretation of experimental results (e.g. Wightman & Green, 1974).

Surprisingly, when white noise is amplitude-modulated, there is no significant change in its spectral content (Rice, 1955 ; Miller & Taylor, 1948) ; the only difference is in the amplitude-time envelope, which reflects the periodicity of the modulating frequency. In 1977 I gave 14 female green treefrogs a choice between unmodulated noise and noise which was amplitude-modulated (about 75%) with a 50 Hz tone. Both signals were electronically filtered so that their spectral envelopes were similar to that of the synthetic call depicted in Fig. 1B. Thirteen of the 14 animals chose the unmodulated sound, thus indicating that H. cinerea detected the difference on the basis of the amplitude-time envelope alone (Table 2). It will be important to generalize this result by presenting a choice between unmodulated noise and noise which is amplitude-modulated at a rate of 300/s. Here I predict that the modulated noise will be preferred provided that the depth of modulation is sufficient. [It may even be necessary to burst the noise on and off 300/s with a duty cycle of 0·5 or less. See Miller and Taylor (1948) and Small (1955) for some relevant psychophysical results.]

Capranica (1976; see also Moffat & Capranica, 1974) has recorded extensively from single auditory neurones in the eighth nerve of H. cinerea from my study area in Georgia. He finds three distinctive populations of neurones: a low-frequency population tuned to frequencies below 500 Hz, with a peak around 300 − 400 Hz; a mid-frequency population tuned to around 600 − 1200 Hz, with a peak around 900 Hz; and a high-frequency population, broadly tuned to around 3200 − 3600 Hz. The absolute thresholds (at their best frequencies) of the high-frequency units are, on the whole, higher than those of auditory neurones in the other two populations. Some neurones in all three populations fire in a phase-locked manner to repetitive clicks, up to click rates of about 300/s. The results of the behavioural experiments relate to these basic observations in two principal ways.

I have argued and presented evidence that some minimum change in the amplitude-time envelope of a sound is required for the detection of waveform periodicity. The actual details of the fine-structure should be irrelevant for mating call recognition. It will be very important to discover if single auditory neurones can follow faithfully the amplitude changes in a stimulus such as the standard call with 300/s beats; the amplitude changes in this stimulus are much less pronounced than in repetitive clicks.

Neurones in both the mid-and high-frequency populations show the ability to follow repetitive clicks, and the behavioural experiments indicated that temporal information can be carried by either frequency band (to which these populations of neurones are tuned). For example, females chose a call with 300/s beats which lacked high-frequency energy altogether (0·8+1·1 kHz; Table 3). Since the high-frequency units (tuned around 3 kHz) are significantly less sensitive than the mid-frequency units (tuned around 1 kHz), then the detection of 300/s periodicity (beats) should occur at a lower SPL when the information is carried by the mid-frequency units. When both sounds were played back at an SPL of 60 dB, females failed to discriminate between the Standard call in which the beats arise from the interaction of the high-frequency components (0·9 +2·7 +3·0 kHz) and the call without beats (0·9 + 3·0 kHz). In 1977, 12 females were given a choice between the call without beats (0·9+3·0 kHz) and a call in which the beats arise from an interaction of the low-frequency components (i.e. 0·9 + 1·2 +3·0 kHz). Eleven of the 12 animals chose the call with beats when both sounds were played back at a 60 dB SPL (Table 3).

The role of the low-frequency (tuned below 500 Hz) units in temporal pattern recognition is uncertain. These neurones respond to beat-frequencies around 300 Hz provided that the components giving rise to the beats are not too high; more specifically, the neurones would probably respond to the beats arising from 0·9 + 1·2 kHz but not to the beats arising from 2·7 + 3·0 kHz (Capranica & Moffat, personal communication).

In any event, the responses of neurones at the level of the eighth nerve merely reflect the possible encoding of the temporal information in the acoustic signal. The extraction of temporal information undoubtedly depends on neural processing at higher levels in the auditory system. The behavioural studies have demonstrated that green treefrogs preferentially respond to sounds with quite subtle differences in temporal patterning. In some cases the failure to discriminate may reflect the resolution limits of the auditory system, but I emphasize that this is merely a hypothesis. Even if the auditory system provides accurate information about the differences in two signals, there is no compelling reason to believe that the female must act on this information ; some temporal differences may be of little or no biological significance to the green treefrog.

The set of temporal specifications for mating-call recognition presented in this paper also provides some insight into the function and evolution of the green treefrog communication system. In the southeastern United States, H. cinerea often breeds in close proximity to other frogs, including treefrogs with rather similar mating calls. Furthermore, males of H. cinerea frequently produce other kinds of distinctive vocalizations besides the mating call. Finally, significant variation occurs in both the spectral and temporal properties of the mating calls of males within and between populations (Oldham & Gerhardt, 1975; Gerhardt, unpublished data). Since the female can select and locate an appropriate mate solely on the basis of the calls he produces, the synthetic call studies provide some information about the sound patterns used for species identification, the differentiation of signals within the repertoire, and sexual selection. I shall now discuss the first two functions; a study of sexual selection is still in progress.

A previous study (Gerhardt, 1974a) indicated that spectral differences alone would probably be sufficient for effective discrimination by female green treefrogs against the calls of two closely related, sympatric species, H. gratiosa (barking treefrog) and H. andersonii (pine barrens treefrog). The results of the present study indicated that fine-temporal differences alone are probably insufficient. Mating calls of H. gratiosa consist mainly of a harmonic series, which follows a short, pulsatile beginning similar to that in H. cinerea calls. The fundamental frequency of the series and hence the waveform periodicity ranges from about 410 − 520 Hz. Since females of H. cinerea failed to discriminate between the standard (300/s) filtered pulse call and a call with a periodicity of 450/s (Table 1), it is unlikely that that difference in waveform periodicity in the natural calls would effectively differentiate the sounds as far as H. cinerea females are concerned. The mating calls of H. andersonii are very similar to those of H. cinerea in their fine-temporal structure (Gerhardt, 1974 b).

Another species of treefrog which often calls in close proximity to H. cinerea is the squirrel treefrog (Hyla squirella). A typical mating call is compared with mating calls of H. cinerea in Fig. 5. As in H. cinerea, the mating call of H. squirella has a bimodal spectrum: the low-frequency peak is found around 1·2 − 1·4 kHz and the broader, high-frequency peak is centred around 3·2 kHz (Gerhardt, 1970). The slightly higher low-frequency peak is the only significant spectral difference, and some small H. cinerea produce calls with low-frequency peaks in this region of the spectrum. There are several differences in the temporal properties of the calls of the two species. Calls of H. squirella tend to be slightly longer in duration, but the ranges of variation in call duration overlap broadly in the two species. Calls of H. squirella are distinctly pulsatile throughout, and the temperature-dependent periodicity of the calls ranges from about 100/s to 130/s between 20 and 26 ° C (Gerhardt, 1970). The difference in the shape of the amplitude-time envelope is particularly evident in a comparison between a typical call of H. squirella (Fig. 5 c) and an atypical call of H. cinerea (Fig. 5 b). In both calls the quasi-periodicity of the waveform is less than 160/s; however, the oscillograms show that the periodicity is very evident in the amplitudetime envelope of the H. squirella call but not in the H. cinerea call. In other words, the H. squirella call is similar to the 100/s filtered-pulse call (Fig. 1d) and the 100/s AM call (Fig. 2c) in that large changes in amplitude with time are obvious. Three females of H. cinerea and 14 females of H. squirella, which were given a choice between H. squirella and atypical H. cinerea calls (the same call which is illustrated in Fig. 5 b and several others produced by the same male), responded to conspecific calls without exception. Furthermore, the results of experiments with green treefrog females indicated clearly that filtered-pulse, amplitude-modulated, and beat stimuli with periodicities of 100/s were less attractive than any other competing stimuli (Tables 1–3). In these experiments, spectral differences were minimal. Thus, I suggest that differences in temporal structure alone are sufficient for females of H. cinerea to identify conspecific males in sites where H. squirella also calls.

cinerea. When a male of H. cinerea touches another male, one or both animals typically produce a series of highly pulsatile sounds, which have been functionally labelled ‘release’ calls. These sounds have a very similar structure to other calls which are produced when the males are not in physical contact (Fig. 6). These ‘pulsed’ calls are usually heard most frequently early in the evening when males are establishing calling positions within the breeding site. Presumably they function to space males (Garton & Brandon, 1975; Oldham & Gerhardt, 1975). (The mating call is also involved in vocal interactions among males, and for this reason Wells (1977) suggests the mating call be designated as the ‘advertisement’ call since it signals the presence of a male to other males as well as to females.) Essentially, both kinds of calls are amplitude-modulated at rates in the order of 50/3 (range 40 − 66/s; mean 48/s; Oldham & Gerhardt, 1975); the modulation appears to be superimposed on the basic 300/s periodicity of the typical mating call. When females were given a choice between the mating calls and ‘pulsed’ calls of the same male, they almost always chose the mating call (Oldham & Gerhardt, 1975). Thus, the demonstration that female green treefrogs discriminated against synthetic calls with pulse-repetition rates of about 50/s was not surprising. I emphasize, however, that pulsed calls are effective in attracting females when mating calls are unavailable. Females chose pulsed calls over the mating calls of H. gratiosa (Oldham & Gerhardt, 1975) and they readily responded to synthetic pulsed calls (except filtered pulses) in the absence of a competing, non-pulsed signal. A more detailed study of the basis of the discrimination between pulsed and non-pulsed calls is currently underway. It is clear at this point that both the depth of modulation and the number of cycles of modulation are important (Gerhardt, unpublished data).

The preference for the mating call may support the idea that females select mates in a chorus on the basis of the quality of their calls. If a female responds to a mating call, there is a much higher probability that she will achieve amplexus with the male whose call she may have assessed than if she responds to a pulsed call, which indicates that at least two males are in close proximity. Another reasonable interpretation is that the female is avoiding the aggressive male behaviour associated with the production of pulsed calls.

I thank Mr G. Williamson, Mr C. Milmine and Ms G. McClung for use of facilities at the Savannah Science Museum; Mr R. Daniel, Ms S. Simon, Dr D. Forester, Dr S. Perrill, Dr J. Rheinlaender, Ms S. Hopkins, Mr C. Seyle and Mr D. Yaeger for field assistance; and Dr R. Capranica, Dr S. Perrill and Ms A. Moffat for helpful comments on the manuscript. I am especially grateful to Dr W. Sherman for the design, construction, and maintenance of the synthetic call-generating system. This research was supported by the Research Council of the University of Missouri, Columbia, the National Science Foundation and a Research Career Development Award from the National Institutes of Health.