The Method Based On Wavelet Transform To Extract The Voiceprint Parameters

In voiceprint recognition system, the ability to extract the precise parameters of voiceprint crucial is important, because it directly affects the recognition rate of the entire system. The traditional MFCC parameters are put forward as the most widely used to mimic the human ear’s auditory characteristics, but it ignores the dynamic characteristics of the speech signal. The wavelet transform will not be bound by short steady hypothesis, and it has good dynamic characteristics, so it is more in line with the hearing sense characteristic of the human ear. Thus the wavelet transform can be applied to the processing of the speech signal, a method is presented for extracting voice parameters based on the wavelet transform(the WPT parameter method).The text independent voice recognition system in order to enhance a more prominent feature of the speaker and overcomes the speech content different effects on the voiceprint parameters extraction, The frame length is adopted to 2560 points in framing stage for increase effective voice. By comparing the simulation WPT parameters and the MFCC parameters of a frame of the speech signal, derived the 16 MFCC parameters, only in front of about seven or eight more obvious value, but the latter value is very small, tends to zero, is not conducive to portray speaker characteristics. In contrast, the 16 WPT parameters relatively large changes, more conducive to describe the characteristics of different speakers. Visible compared to the MFCC parameters, the WPT parameters are better represent the dynamic nature of the speech feature parameters over time, which will help improve the recognition rate. To further verify the performance, then by comparing the simulation 16-dimensional MFCC parameters and the 16-dimensional WPT parameter of former ten frames of the speech signal. Apparently found by comparing the simulation map with the MFCC parameters, the shape between frames, the WPT parameters are more similar to the characteristics that means closer. Therefore, the WPT parameters are better than the MFCC parameters for speaker recognition.On MATLAB platform, the vector quantization(VQ) is combined with the speaker recognition experiment system to verify the extracted parameters respectively recognition rate, and through the comparison of commonly used db3、db4、db6、coif3 wavelet function to choose the best basis. Experimental results show that frame length within 2560 points is higher and improves computing speed in comparison with common 256 points of the frame length. The optimal base coif3 wavelet function is taken as voiceprint parameter extraction. The new WPT parameters recognition rate is better than the traditional MFCC parameters.