Speaker verification system using hybrid model with pitch detection by wavelets

For speaker-verification, we design a hybrid-recognizer composed of conventional methods (such as HMM, MLP, and DTW) as pre-recognizer and pitch-detection method through wavelet transforms as post-recognizer. Conventional methods have respectively 3 classes of feature vectors; LSF, cepstrum, and filter bank. The pitch carries information about speaker identification, which is obtained by wavelet analysis. Wavelet analysis has advantages over traditional Fourier methods in analyzing physical situations where a signal contains discontinuities and sharp spikes, etc. The pitches in speech analysis are composed of impulses, so that a wavelet transform for detection of the pitch period can be used. This system can analyze pitch period under various noisy environments.