Abstract:

This paper introduces a feature extraction method based on a new wavelet filter. At first, the new wavelet’s theory is introduced. Then， the new wavelet filter is designed according to the concept of human critical frequency band， and the scale parameter which the new wavelet filter need is given. The SDA9000 is used for spectral analysis, the LSP is applied for FPGA hardware simulation. The SNN (Synergetic Neural Networks) is used in train and recognition, and the Gauss wavelet filter is used to compare with the new wavelet filter. The characteristics of numerical and application for the methods are illustrated by using PC simulation of Matlab GUI. After the analysis of the spectrogram and the recognition result, it is found that the new wavelet filter has higher recognition rate and better robustness than traditional feature.

Key words: speech recognition, auditory model, auditory filter, critical bands, wavelet filter