The role of the auditory periphery in the categorization of stop consonants
R.I. Damper · The Journal of the Acoustical Society of America · 1998
A proper understanding of speech perception requires an understanding of the staged restructuring of information that occurs as an acoustic stimulus is transformed by the auditory system into a phonetic percept. It is now firmly established that a physiologically faithful computer simulation of peripheral processing feeding into a trainable artificial neural network (ANN) is capable of reproducing the important aspects of the categorization of initial stop consonants into voiced and unvoiced classes. Correct behavior is found to be insensitive to the ANN architecture and a precise training scheme used, suggesting that such categorization is very basic to the repertoire of auditory processing strategies. Unlike a human or animal listener, a computational model can be easily manipulated and interrogated to discover the underlying basis of category formation. In particular, the physiological-based simulation of the auditory periphery can be dramatically simplified and the effect on categorization observed. It is found that proper modeling of frequency scaling and neural adaptation are essential to correct simulation of the well-known shift of phoneme boundary with place of articulation for stop consonants. The role of the auditory periphery in this case seems to be to emphasize the region of first formant information around the time of voicing onset.