Excitation-synchronous modeling of voiced speech

A new modeling technique for voiced speech is introduced. Salient features are detailed modeling of speech waveforms and the use of improved parameter estimation techniques. The ideas of pitch-synchronous analysis are extended to make two subintervals synchronous with regions of approximately closed and approximately open glottis. Two LPC models are used in each pitch period, and the model parameters are changed at estimated times of transition from open-to-closed and closed-to-open glottis. The excitation is provided by changing initial conditions at these transition instants. Experiments with real, connected speech indicate that the speech waveforms can be accurately represented using the analysis-synthesis approach presented here.