MIXED-PHASE SPEECH MODELING AND FORMANT ESTIMATION , USING DIFFERENTIAL PHASE SPECTRUMS

This paper introduces a new speech model, termed as the mixed-phase model, based on the assumption that the speech signal is produced by convolution of a maximum phase glottal excitation signal with a minimum phase vocal tract filter impulse response. The glottal excitation signal is assumed to be an anti-causal stable signal and the vocal tract filter is assumed to be causal and stable. For estimating resonances of the maximum phase signal (source) and the minimum phase filter (vocal tract filter), use of differential phase spectrums of z-transforms is proposed.