Research on Extracting Algorithm of Speech Eigenvalue Based on Wavelet Packet Transform and Gammatone Filter

Whether the performance of speech system is good or not is closely related to its characteristic parameters. An excellent feature extraction method is a key step in forming a good recognition system. A good feature extraction method can improve the design and performance of the acoustic classifier. In this paper, a feature extraction algorithm based on human-ear perceptual characteristics is presented, which differs from the traditional Mel Frequency Cepstrum Coefficients (MFCC). This paper presents and studies the method of extracting feature parameters from Wavelet packet transform applied to bark-scale filter, and the feature extraction method based on Gammatone filter. Finally, it is proved that the Wavelet packet transform and the feature extraction method based on Gammatone filter can greatly improve the recognition rate through simulation and experiment.