A novel approach to the estimation of voice source and vocal tract parameters from speech signals
Wen Ding, Hideki Kasuya · 2002
The paper presents a novel adaptive pitch-synchronous analysis method for simultaneous estimation of voice source and vocal tract (formant/antiformant) parameters from the speech signal. The method uses a parametric Rosenberg-Klatt model to generate a glottal waveform and an autoregressive with exogenous input (ARX) model for representing the speech production process. The time-varying coefficients of the model are estimated with an adaptive algorithm based on a Kalman filter, while the parameters of the Rosenberg-Klatt model are optimized using the simulated annealing method. In addition, a new hybrid error criterion is used to optimize the glottal opening instant. Furthermore, in order to estimate the fundamental period parameter T/sub 0/, it is defined as two successive glottal closure instants, and is estimated automatically based on the obtained differentiated glottal waveform. Experiments using two-channel speech signals (speech and electroglottograph (EGG) signal) and continuous speech show a good estimation performance.