A speech analysis technique based on temporal and frequency masking properties of the auditory system

A novel approach about an auditory speech analysis technique is presented in this paper. This analysis is based on an atomic decomposition of speech signal into a family of gammachirp functions. An auditory model is used to simulate the outer and middle ear filtering, the spectral behavior of the cochlea and the time domain processing of the auditory system. A spectrographic representation is finally given as an auditory image, which will be transmitted to the brain. The obtained spectrograms highlight at the same time, the harmonics of the pitch and the structure of formants. Furthermore, they present a good distinction between fricatives, which favors these auditory spectrograms versus classical short-time Fourier spectrograms.