NEURAL COMPUTATIONS FOR SOUND PATTERN-RECOGNITION - EVIDENCE FOR SUMMATION OF AN ARRAY OF FREQUENCY FILTERS IN AN ECHOLOCATING BAT

Microchiropteran bats use an auditory sonar system for orientation and prey capture. Many bats use highly structured constant-frequency (CF) and frequency-modulated (FM) sonar orientation signals. Mechanisms for sound pattern recognition are important for the perception of these and other types of auditory signals. The processing and recognition of FM sound components appears to be important for certain complex perceptual tasks, including target distance perception. I have conducted behavioral studies using artificial echoes to simulate the conditions of a bat flying toward a target. An innate vocalization response of the bat to the simulated approaching target was used to assess the ability of the bat to analyze the structure of and extract distance information from different types of synthetic FM sound patterns. The bat's performance depended on the structure of the artificial echo. The pattern recognition performance of the bats was similar when they were presented with either a naturally structured artificial CF/FM echo or an artificial CF/FM echo containing an FM component consisting of a series of pure tone steps. The ability of the bats to recognize appropriately the structure of an FM signal constructed from a sequence of pure tones depended on the number of pure tone steps in the series. Noctilio was able to recognize FM sound patterns containing 99 or greater pure tone steps. The minimum required number of pure tone steps could be distributed over different frequency ranges. The bats were able to resolve individual tone steps in the series that were separated by at least 100 Hz. These results provide behavioral evidence suggesting the existence of an array of neurons, each of which responds to a specific frequency or narrow range of frequencies in the FM sweep. A critical computation step seems to involve neural summation of information from this array of frequency filters.