On temporal smoothing and speech perception

Abstract

Neurophysiological and psychophysical studies provide evidence for the existence of temporal integration mechanisms in the auditory system. These may be viewed as low-pass filters, parametrized by their cutoff frequencies. It is of interest to specify these cutoffs, particularly for tasks germane to the effect of temporal smoothing on speech quality, within the context of speech-coding strategies (e.g., what is the smallest number of bits with which to represent temporally smoothed speech that is perceptually indistinguishable from the original?). In order to answer such questions, speech-processing rules have been derived so as to create synthetic speech capable of generating model auditory-nerve (AN) firing patterns temporally smoothed in a manner commensurate with that observed in neurophysiological studies of the AN (e.g., at low characteristic frequencies, neural discharges of AN fibers are phase locked to the underlying driving critical-band signal, i.e., synchrony is maintained; at high CFs, temporal information is preserved by the instantaneous average rate of the neural firing pattern, which is related to the temporal envelope of the underlying driving signal). This presentation describes the processing rules and the results of psychophysical experiments performed to measure the cutoff frequencies of these filters within the context of speech quality.