Research And Application Based On Wavelet Transform For Speaker Recognition

The development of Wavelet transform provides new processing method and technology for speech signal, it makes speech processing technology made great progress. Speaker recognition is extracted speaker the phonetic features of the speaker's identity confirmation or illegible. Speaker recognition is an important research field, through the speech signal effectively extract personal characteristics information for identity recognition.In the speaker recognition system, the most important thing is being able to extract from phonetics fragments represent speaker unique features of stable parameter. Based on the short time Fourier transform showed many deficiencies in the short smoothly, Wavelet analysis includes integral wavelet transform and wavelet series. As a multiresolution representation, wavelet transform has different resolutions on time-frequency domain. The basic function of wavelet transform (wavelet) is built by dilating and translating the window function. The theoretical and experimental analysis result shows wavelet transform to extract the speaker recognition characteristic parameters obtained the certain effect, than the traditional methods of analysis are definitely better recognition properties.Speech recognition rate of increase need speech signal pretreatment, the paper also use wavelet transform knowledge of the recognition of the speaker speech endpoint detection and noise reduction processing, in order to ensure that the extract characteristic parameters of accurate. Aiming at the characteristics of wavelet analysis, skip signal reconstruction after pretreatment, will directly component to enter the step processing, simpling the process and improving precision.