Vocal tract excitation during voicing and its relation to optimal short-time frequency analysis of speech

Abstract

The physics of vocal tract excitation during voicing is presented with emphasis on the transient nature of the excitation. The vocal tract is excited by a delta function like impulse at the time of the closing of the glottis. After reflecting from the mouth opening the impulse excites standing modes in the vocal tract. Knowledge of this time sequence leads to some conclusions about the best interval in which to analyze the signal. The optimal time window for obtaining articulator positional information includes only the ringing portion of the time waveform and exclude the excitation pulse. Therefore this optimal frequency analysis is a type of pitch synchronous analysis. Detailed comparisons with other types of short-time frequency analysis are made, including the sound spectrograph, and autocorrelation method of linear predictive coding. Pitch synchronous analysis is shown to be closely related to broadband sound spectrograms, while autocorrelation linear predictive spectra are related to narrow-band sound spectrograms, in the case of optimal spectra, the resonant frequencies of the vocal tract are so clearly defined that formants may be easily obtained by peak picking. Since the formants are obtained each pitch period, rapid transitions are followed in greatest possible detail.