Neural coding of complex sounds: speech

The studies that we have reviewed here demonstrate that a temporal-place code can represent fine details in the spectra of vowels and stop-consonants. These features include formant frequencies, formant-frequency transitions, and pitch. On the other hand, such a phase locking-based representation may have difficulty with certain fricative consonants. Detailed format structure of vowels is present in a rate-place code at moderate stimulus levels, but is maintained at high levels only in the small population of low-spontaneous-rate fibers. A rate-place code preserves the formant structure at high stimulus levels better for stop-consonants than for vowels. Formant structure of some fricatives may be represented in a place-rate code in cases where a temporal-place code fails. Voice pitch is well preserved in a temporal code but not in a rate-place code. From this summary we must consider the possibility that the central nervous system utilizes both rate-place and temporal-place information in processing all of the relevant features of speech. This review points out a number of issues that currently confronts us in the coding of complex stimuli. We conclude by briefly summarizing these issues. First, we must attempt to clarify the role of the low-spontaneous-rate, high-threshold fibers in the representation of speech at high-stimulus levels. A second major question that must be addressed is the role of the cochlear efferents in the peripheral encoding of speech, especially in the presence of background noise. Closely related to the role of the efferents may be the role of the middle ear muscles. Finally, if we are to seriously consider a temporal-based code for speech, we must develop quantitative hypotheses about how spectral information might be extracted from temporal patterns in the central nervous system with real neural "hardware."