A feature extraction method based on combined wavelets filter in speech recognition

This paper used wavelet theory in noise-robust feature extraction of speech recognition and introduced two kinds of feature extraction methods based on Gauss wavelet filter and combined wavelets filter. The Gauss wavelet filter and combined wavelets filter with critical frequency bands are obtained by studying human auditory characteristic. Wavelet has flexible characteristic in choosing frequency, the key is making certain the scale parameter. This paper studied the choosing method of scale parameter in designing the two kinds of wavelet filter. The methods used new feature and original feature were simulated. The RBF neural net is used in training and recognition course. The results showed that new feature had higher recognition rate and better robustness than traditional feature.