A theory for an acoustic perception system

Abstract

A theory of acoustic perception based on extracting meaning from a signal by direct analysis of waveform features is presented. This approach avoids the theoretical difficulties that are inherent in Fourier or any other transform method. At each waveform zero, parameters in time, amplitude, and space are measured and stored in a delay line. Using a hierarchy of time-parallel arrays, this information is instantaneously parsed into sequences that characterize the various environmental acoustic sources. The system described emphasizes, but is not limited to, speech perception. Examples demonstrate real-time direction finding, sorting speakers by direction, and automatic phonetic segmentation and labeling.<>