Spoken word recognition using vocal tract shapes
H. Kinugasa, Hiroyuki Kamata, Yoshihisa Ishida · 2002
The estimation method of vocal tract shapes using an adaptive inverse filter for the discrete time signal has been proposed by T. Nakajima et al. (1978). In the present work, the authors use the adaptive inverse filter to autocorrelation coefficients to reduce the computation time. The estimated vocal tract shapes are used for the recognition experiment as input parameters. Speaker-independent word recognition results demonstrate the effectiveness of the method. In the experiment, dynamic programming matching is used. Experimental results show that the recognition rate using vocal tract shapes is higher than when using LPC (linear predictive coding) spectra.>