19 th INTERNATIONAL CONGRESS ON ACOUSTICS MADRID, 2-7 SEPTEMBER 2007

Transcription

1 19 th INTERNATIONAL CONGRESS ON ACOUSTICS MADRID, 2-7 SEPTEMBER 2007 TEMPORAL ORDER DISCRIMINATION BY A BOTTLENOSE DOLPHIN IS NOT AFFECTED BY STIMULUS FREQUENCY SPECTRUM VARIATION. PACS: Lb Zaslavski Gennadi University authority for applied research, RAMOT, Tel-Aviv, Israel; zaslg@bezeqint.net ABSTRACT Bottlenose dolphins are able to discriminate a small click followed by a large click from a timereversed pair of the same clicks at interclick intervals as small as 4-10 µs. The dolphins' ability to discriminate variety of time-reversed signals with identical energy spectra points to extraordinary auditory temporal acuity. In this paper, time domain discrimination of the double clicks is once again weighed against frequency domain discrimination. The Black Sea bottlenose dolphin was required to discriminate time-reversed double clicks with variable frequency spectra. The results show that variation of the double click frequency spectra does not affect the bottlenose dolphin's ability to discriminate the temporal order of a small and large click. INTRODUCTION A 300-µs time interval is widely considered to be a fundamental constant of the bottlenose dolphin auditory system that defined both time and frequency analysis of brief signals. As long as interval between clicks is shorter than around 300 µs, the pair of clicks is believed to fuse for the dolphin into a single auditory image [1-4]. However, there are numerous experimental facts which are obviously at variance with the 300-µs critical interval concept. Among many other contradictions is the bottlenose dolphins' ability to discriminate variety of brief signals with identical energy spectra and durations much shorter than the critical interval of 300 µs [5-10]. Three independent studies [5, 7 and 11] found that both the Atlantic and the Black Sea bottlenose dolphins are capable of discriminating the time-reversed double clicks with identical interclick intervals and energy spectra (Fig. 1). 200 µs 25 µs A B 200 µs C 25 µs Figure 1.-Time-reversed double clicks and their STFT spectrograms generated using 800-µs (300 µs at a 3-dB level) Hanning window and 10-µs time increment. Interclick intervals were 200 and 25 µs. The first-to-second click amplitude ratio is 14 db. (Computer simulation) To somehow account for bottlenose dolphins' ability to discriminate time-reversed double clicks with interclick intervals as small as short as 10 µs [2] (50 µs interval for the Atlantic bottlenose dolphin [11]), Dubrovskiy [2] suggested that within the 300-µs critical interval both the time and frequency analysis are valid. However, such a suggestion virtually denied the critical interval concept altogether because by definition within the critical interval the first and second highlights are fused into a single auditory image.

2 Theoretically, the double clicks can be identified by short-time frequency spectrum as well as by temporal order of a small and large click. Johnson et al. [11] assumed that the dolphin could not temporally resolve the clicks separated by a 200-µs interclick interval and considered possible frequency domain cues. Windowed short-time Fourier Transform (STFT) analysis was used to generate the difference in the frequency spectra of a direct and a time-reversed double click. A chi-square window with 90 % decay by 300 µs was used for a short-time spectral analysis. The chi-square window simulated a frequency analysis with the bandpass auditory filters characterized by a transient response with a rapid rise and exponential decay. The size of the window was chosen based on the dolphin s critical interval of 300 µs [1, 2] and the dolphin s integration time of 265 µs [4]. For a 200-µs interval and 10-dB amplitude difference between the first and second clicks, the 300-µs chi-square window strongly enlarged the ripples for a direct double click and diminished the spectrum rippling for a reversed double click. In fact, a symmetrical analysis window, for example Hanning window, should also produce different short-time spectra for a direct and time-reversed double click if interclick interval is comparable with the window duration (Fig. 1B). However, a 300-µs analysis window could hardly produce distinguishable differences in the short-time spectra of the time-reversed double clicks for a much shorter interval between the first and second clicks of 25 µs (Fig. 1C) or even for a shorter threshold interclick interval of 4 µs found for the Black Sea bottlenose dolphin [12]. The Black Sea bottlenose dolphins discriminated the time-reversed double clicks as long as a first-to-second click ratio was bigger than 0.5 db and smaller than around 30 db [8]. At a very small amplitude difference between the clicks, the short time spectrum of the double clicks appears to be almost identical for long interclick intervals as well as for short intervals (Fig. 2). Nevertheless, the dolphin discriminated the double clicks with amplitude difference between the first and second clicks as small as 0.92 db at 100 % correct level. 200 µs 25 µs A B C 200 µs 25 µs Figure 2.-Time-reversed double clicks and their STFT spectrograms generated using 800-µs (300 µs at 3-dB level) Hanning window and 10-µs time increment. Interclick intervals are 200 and 25 µs. The first-to-second click amplitude ratio is 0.92 db. (Computer simulation) In order to discriminate the time reversed double clicks at very short interclick intervals, bottlenose dolphins appeared to use the auditory filter centered at one of troughs in the energy spectra of the double clicks and apparently discriminated the time domain waveforms of the auditory filter reactions to a direct and reversed double click [8-10]. A trough in the energy spectrum of a double click is associated with rapid (and opposite in a direct and reversed double click) change in the phase spectrum, which causes significant difference in the filter reactions. On the other hand, as long as stimuli have different time waveforms, the short-time spectra of the stimuli are different as well even if the stimulus energy spectra are identical. Therefore, as long as otherwise has been proven, a short-time spectrum should be considered as a possible cue for discrimination even when the differences in short-time spectra are very small. A straightforward way to eliminate short-time spectrum differences is to mask the double clicks with continuous broadband noise. Because the short-time spectrum generation involves stimuli integration within an analysis window, masking of the time-reversed double click with continuous noise should affect short-time spectrum stronger than it affects the time waveform of the double clicks. Threshold signal-to-noise ratio may indicate the cue used by the dolphin for discrimination. Short-time frequency spectra of the time reversed double clicks can be also 2

3 distorted by random variation of interclick intervals. This paper reports on a bottlenose dolphin response to the frequency spectrum variation of the time-reversed double clicks. METHOD The subject was the Black Sea bottlenose dolphin (Tursiops truncatus). Experiments were conducted in a m concrete pool. The two-response forced-choice procedure was used. A vertical net partition between two transducers set a minimum distance of 5 m, from which the dolphin was forced to make his choice. Signals were transmitted simultaneously through transducers situated at 1m depth and 3 m from each other. Prior to stimuli presentation, the dolphin positioned itself at the far (from the transducers) end of the partition. 1.5-cm piezoelectric spheres were used as transducers. The maximum of the transducer transmitting response was at khz. Standard analog electronic equipment was used to produce stimuli. Masking noise was mixed with the stimuli and transmitted through the same transducers. A two-response forced-choice procedure was used. Periodic stimuli were presented to a dolphin at a repetition rate of 2 to 4 stimuli per second. Threshold measurements were made using a method of constant stimuli. Signals were presented in 10-trials blocks with the same signal parameters. The dolphins performed 250 to 400 trials per session. The threshold values were estimated at a 75% correct response level. In the first series of experiments the dolphin was requested to discriminate the time-reversed double clicks masked with continuous broadband noise. Threshold signal-to-noise ratio was measured as a function of the interclick interval. The results were used to examine how masking of the short-time spectra depended on the length of analysis window. The longer the analysis (integration) window, the stronger the noise masks the short-time spectrum of the double clicks. If the short-time spectra are completely masked (become indistinguishable) long before the dolphin fails to discriminate the double clicks, it would signify that this particular window is much longer than the actual auditory analysis window used by the dolphin. Another way to interfere with the frequency spectra discrimination was to randomly vary the frequency spectra of stimuli. The dolphin was required to discriminate the time-reversed double clicks with interclick intervals randomly varied during a trial (Fig. 3). A B C Fig. 3. Superimposed waveforms of several consecutive time reversed double clicks with varied interclick intervals but identical energy spectra for any simultaneously presented double clicks (A) and with varied and different interclick intervals and energy spectra (B and C). Bottom frames are energy spectra of the double clicks shown in the middle row. The first-to-second click amplitude ratio is 0.92 db. Interclick intervals were changed either synchronously (Fig. 3A), so that at every simultaneous presentation of the time-reversed double clicks their energy spectra were identical, or within different margins for a direct and reversed double click (Fig. 3B and 3C). Moreover, in different trials these margins for a direct double clicks could be larger or smaller than for a reversed double click (Fig. 3B or 3C). Amplitude difference between the first and second click was just 3

4 0.92 db. Because the amplitude difference between the clicks was very small, the double click energy spectra were rippled with deep troughs so that even a small variation of interclick intervals led to significant variation of the frequency spectra (Fig. 3, bottom frames). Additionally, the double clicks with variable intervals were masked with 140-µs broadband noise pulses (Fig. 4). Threshold signal-to-noise ratio was measured as a function of a median interclick interval. Interval deviation from a median interval was around 50%. S/N=32 db A B C 25 µs 60 µs 25 µs without noise S/N=32 db S/N=22 S/N=32 db Figure 4. (A)-Superimposed waveforms of 10 consecutive time-reversed double clicks masked with 140 µs broadband noise pulses. (B and C)- STFT spectrograms generated for two consecutive pairs of double clicks using 800-µs (300-µs at 3-dB level) Hanning analysis window. Signal-to-noise ratio is around 3 db above the threshold. Bottom frames are superimposed energy spectra of the double clicks for interclick interval of 25 and 60 µs in quiet (A) and masked with the noise pulse (B and C). RESULTS Simultaneous masking of the double clicks with continuous broadband noise Threshold signal-to-noise ratios were found to be around 18 db for tested interclick intervals from 25 to 160 µs (Fig. 5). Threshold signal-to-noise ratio (db) Interclick interval (ms) Figure. 5.-Threshold signal-to-noise ratio (peak-to-peak value of a smaller click to a root-mean square value of the continuous noise at the acoustical side of transducers) as a function of the interclick interval. Amplitude difference between the first and second clicks was 14 db (Fig. 1). Even for a signal-to-noise ratio as high as 32 db (Fig. 6), the short-time energy spectra are distorted by the noise for all but the shortest analysis window of 100 µs (35 µs at 3-dB level). Given the fact that the double clicks have identical energy spectra and very small differences in short-time spectra, at least for short interclick intervals, the dolphin had no chance to distinguish 4

5 the short-time spectra even at the signal-to-noise ratio as high as 10 db above the threshold. 10 db above the threshold means that the noise level should be increased by 10 db before it disrupted the double click discrimination. For signal-to-noise ratios near the threshold (S/N=22 db), integration of the noise even with a 100-µs analysis window completely masks the short-time spectra. A shorter analysis window, for example 30 µs, is not long enough to integrate the first and second clicks separated by 25 µs (Fig. 6) and generate any differences in the short-time spectra of the double clicks. However, a 30-µs window would be short enough to enable the dolphin to discriminate the temporal order of a small and large click. 25 µs 25 µs 25 µs 25 µs A B C D Figure 6.- Time-reversed double clicks masked with continuous broadband noise and their energy spectra (A). STFT spectrograms were generated for the double clicks using 100 µs (B), 200 µs (C) and 400 µs (D) Hanning windows (35, 75 and 150 µs at 3-dB level, respectively). Signal-to-noise ratio is 32 db (10-12 db above the threshold). Bottom frames in (B D) are STFT spectrograms for the same double click as the middle row but in quiet (without noise). Contrary to the short-time spectra, the time waveforms of the double clicks are much less affected by the noise even at signal-to-noise ratios near the threshold, so the double clicks still can be discriminated based on temporal order of a small and large click. The dolphin apparently discriminated the temporal order of a small and large click despite that the double clicks had practically random frequency spectra. The threshold signal-to-noise ratio of around 20 db (Fig. 5) is a detection threshold for a smaller click in continuous broadband noise. As long as the dolphin detected a smaller click in the noise, he was able to discriminate temporal order of the clicks. Random variation of intervals between a small and large click The bottlenose dolphin readily discriminated the double clicks with variable interclick intervals at any combination of the interclick intervals (Fig. 3A, 3B or 3C). 100% correct discrimination of the double clicks was observed for median intervals from 12 to at least 160 µs and interval deviation from a median interval of 20 to 50 %. The dolphin also had no problem in discriminating the temporal order difference between the double clicks with very different median interclick intervals, for example 20 and 50 µs. The dolphin is able to discriminate around 10% difference in interclick intervals [12]. Therefore to discriminate temporal order of a small and large click, he had to ignore well recognizable differences in interclick intervals and of course differences in the double click energy spectra. In fact, it was easy to make the dolphin discriminate differences in interclick intervals between the same double clicks and thus ignore the differences in the click amplitude. At least near threshold signal-to-noise ratios (Fig. 7), the noise pulse completely masked very small regular differences in the short-time energy spectra of the double clicks (compare Fig. 2 and Fig. 4). Interclick interval variation combined with the simultaneous masking of the double clicks with the noise pulse made absolutely impossible for the dolphin to discriminate the shorttime spectra of the double clicks. 5

6 Threshold signal-tonoise ratio (db) Median interclick interval (ms) Figure 7.-Threshold signal-to-noise ratio as a function of a median interclick interval. Interval deviation from the median interval was 50%. The first-to-second click ratio was 0.92 db. Duration of the masking noise pulse was 140 µs (Fig. 4). Threshold peak-to-peak values of a smaller click to rms value of the noise (within the pulse) were measured. For the double clicks with a very small amplitude difference between the first and second clicks (Fig. 2 and 4), the threshold signal-to-noise ratio was reached when the noise level became high enough to mask the amplitude difference between the clicks. For a larger amplitude difference between the first and second clicks of 6 db, the threshold signal-to-noise ratio was found to be around 20 db. CONCLUSIONS Random variation of the frequency spectra of the time-reversed double clicks did not affect the bottlenose dolphin's ability to discriminate temporal order of a small and large click even when interclick intervals were as small as µs. Interclick interval variation combined with the simultaneous masking of the double clicks with the noise pulse made absolutely impossible for the dolphin to discriminate short-time spectra of the double clicks. As long as there was an amplitude difference between the first and second clicks, the dolphin appeared to be unconcerned with the double click frequency spectra. The results show that the double click discrimination by the bottlenose dolphin is purely based on the auditory analysis of temporal order of a small and a large click. References: [1] V.A Velmin and N.A. Dubrovskiy: The critical interval of active hearing in dolphins, Sov. Phys. Acoust., 2, (1976) [2] N.A. Dubrovskiy: On the two auditory subsystems of dolphins: in Sensory Ability of Cetaceans (J. A. Thomas and R. Kastelein, eds), Plenum Press; New York, (1990) [3] P.W.B. Moore, R.W. Hall, W.A Friedl and P.E Nachtigall: The critical interval in dolphin echolocation: what is it? J. Acoust. Soc. Am, 76 (1984) [4] W.W.L. Au, P.W.B. Moore and D.A. Pawloski: Detection of complex echoes in noise by an echolocating dolphin, J. Acoust. Soc. Am., 83, (1988) [5] N.A. Dubrovskiy, P.S. Krasnov and A.A. Titov: Auditory discrimination of acoustic stimuli with different phase structures in a bottlenose dolphin, in Marine Mammals, (1978) [6] G. L. Zaslavskiy: Double-click representation in the dolphin auditory system, in Proceedings of the Institute of Acoustics Symposium on Underwater Bio-Sonar and Bioacoustics, v 19, 9, Loughborough University, United Kingdom, (1997) [7] G. L. Zaslavskiy: Temporal order discrimination in the dolphin, in Proceedings of the Institute of Acoustics Symposium on Underwater Bio-Sonar and Bioacoustics, v 23, 4, Loughborough University, United Kingdom, (2001) [8] G. L. Zaslavskiy: Click discrimination in the dolphin, in Proceedings of the Institute of Acoustics Symposium on Underwater Bio-Sonar and Bioacoustics, v 23, 4, Loughborough University, United Kingdom, (2001) [9] G. L. Zaslavskiy: Discrimination of the signals with identical energy spectra by the bottlenose dolphin, In Oceans 2003, San Diego, California, CD proceedings, (2003) [10] G. L. Zaslavskiy: Time domain discrimination of the time-reversed signals by the Black Sea bottlenose dolphin, In Proceedings of the Eight European Conference on Underwater Acoustics, Carvoeiro, Portugal, (2006) [11] R.A. Johnson, P.W.B. Moore, M.W. Stoermer, J.L. Pawloski, and L.C. Anderson: Temporal order discrimination within the dolphin critical interval, in Animal Sonar: Processes and Performance (P.E. Nachtigall and P.W.B. Moore, eds), Plenum Press; New York, (1988) [12] G. L. Zaslavskiy: Differences between the auditory system of humans and bottlenose dolphins, in Advances in Bioacoustics 2, Razprave IV. Razreda sazu XLII-3, Ljubljana (2006)