A proposed speaker recognition algorithm based upon wavelets packet transform

摘要

In this paper a new speaker recognition algorithm depending upon the WAVELET PACKET TRANSFORM (WPT) is proposed. The WPT works as an analysis tool for the decomposition of the speech signal to evaluate the information carried by the signal. It is based upon the speech words, in which the recognition process is depending on phoneme by phoneme comparision. Each word is segmented into phonemes which is considered as a number of frames of equal length, next for each frame the (WPT) is computed up to the third level (and may be higher if a very high resolution or accuracy is wont to be achieved). The proposed algorithm will next process the WPT coefficients for the training and the recognition phases. According to the given procedure a perfect recognition rate was achieved for ten speakers.