The current data show that for vampire bats, prey-generated breathing sounds could provide a reliable cue for recognizing prey individuals: during the relatively long time a vampire feeds on a prey animal, it can memorize the prey's breathing sounds and use this information to find the same prey on the following night.
Breathing sounds are typically faint. The sounds we recorded from human subjects ranged between 25 and 35 dB SPL. This gives rise to the question: over what distance could breathing sounds be perceived and analyzed by vampire bats? Considering the absolute thresholds (cf. Fig. ), the frequency region most likely to be used is around 15 kHz, where thresholds are as low as 0 dB SPL. In this frequency region, atmospheric attenuation is around 0.5 dB/m. Thus, in the (unlikely) absence of any masking sounds, detection of breathing sounds could work over several tens of meters. In the presence of natural masking sounds, however, the effective detection distance will depend on the level and spatial distribution of the masking sound sources.
While it is unlikely that prey recognition relies exclusively on breathing sounds [5
], these sounds potentially have high individual significance: vocalizations are generated by the vocal cords and filtered through the vocal tract. Both the pattern of vocal-cord vibrations and the filtering are highly individual and this supports our recognition of individual voices. While breathing sounds are unvoiced, and thus do not excite the vocal cords, they will also pass the same vocal and nasal tract and may thus also mediate individually specific information. However, the sounds used in this study were emitted through the nose. It remains to be investigated to what extent the nasal acoustic tract filters breathing sounds in a similarly characteristic way. An early study confirmed that, at least for partially voiced sounds such as consonant-vowel combinations, speaker recognition is feasible on the basis of nasal co-articulation [10
]. However, the current simulations based on the breathing-sound power spectra, and the human-psychophysical experiments, suggest that speaker recognition is difficult with purely unvoiced sounds. If nasal-tract filtering were individually specific, the resulting spectral features would result in a correct breathing-sound association in the power-spectrum simulation. The simulation results in Fig. show that this is not the case. Also, the failure of the instructed human listeners to associate the breathing sounds recorded under physical strain argues against the use of power-spectrum information for breathing-sound recognition, at least in the audio frequency range below 20 kHz. Note that human listeners are very sensitive to changes in the spectral composition of broadband stimuli [11
Qualitatively correct predictions of the vampire-bats' performance, even under the most difficult experimental condition where the test sounds had been recorded under physical strain, could be obtained with a refined simulation approach: first, the sounds were filtered to match the vampire-bat audiogram; and second, a simulation paradigm was designed that allows whichever sound parameter yields the strongest predictions to be exploited.