Research of the auditory feature extraction algorithm based on Gammatone filter bank

摘要

The human auditory system has the ability of analyzing and recognizing signals. A good algorithm of auditory feature extraction can improve the performance of the speaker recognition system. The algorithm embarks from the simulation of human auditory perspective, and first the Gammatone filter is improved, and the improved filter(Gammatone-4 filter) can be used to more better simulate the human cochlea of auditory model, then introduces the equal loudness curve to pre-emphasis. Meanwhile, preliminarily extracted the GFCC. Finally, after the half-liter sine function cepstrum lifting technology to extract the speaker's voice signal characteristic parameters NewGFCC. Experiments show that the algorithm has obvious improvement in recognition rate and robustness of speaker recognition.