ACOUSTIC CORRELATES OF PHONETIC PERCEPTION

点击翻译按钮获取中文摘要

DESCRIPTION: Numerous practical and theoretical problems could be addressed if we had a better understanding of the auditory mechanisms underlying phonetic recognition. Among the practical applications of this knowledge are: (1) the improvement of speech synthesis devices, (2) the development of robust speech recognition devices, (3) the development of acoustically based training devices for hearing-impaired speakers, and (4) improvement in Cochlear-implant signal processors. The proposed experiments fall into three major categories. One set of experiments follows in a rather direct way from vowel perception studies conducted during the previous grant period. These experiments address issues such as the role of dynamic spectral cues and voice fundamental frequency in vowel perception. A second series of experiments address more fundamental issues regarding the spectral representations that control phonetic quality. A major goal of these experiments is to test the validity of a method of representing speech that was developed during the previous grant period. The "Masked Peak Representation" (MPR) was developed as an alternative to both formant representations and whole spectrum models. The MPR involves a series of spectral manipulations that are designed to remove aspects of the spectrum that do not appear to have a strong influence on phonetic quality, while retaining those features that are most relevant to phonetic quality judgments. The MPR will be evaluated with: (1) an experiment comparing MPR-based predictions of perceived phonetic distance with those of a more traditional auditory model, (2) speech recognition tests that use a Hidden Markov Model to map sequences of MPR spectra onto words or phonetic segments, and (3) listening tests with speech resynthesized from MPR spectra. A third set of studies is aimed at modeling the low-level auditory mechanisms that are responsible for spectrum analysis. The goal of this work is to evaluate a model of spectrum analysis that is carried out by the central auditory system rather than the auditory periphery. A software simulation of the model will be developed in an effort to determine the extent to which the central-spectrum model can account for a broad range of findings from the auditory psychophysics literature. Experiments are also proposed that address the implications of this model for vowel perception and for the representation of pitch and periodicity.