Speaker Recognition Research Based On Cochlear Filter Cepstrum Coefficient

Among the biometrics,speaker recognition technology is a common recognition technology.Meanwhile,many researches and applications of speaker recognition technology have been paid more and more attention.Speaker recognition technology is to let the machine simulate the characteristics of the human ear to recognize and confirm of the speaker.As an auditory organ,the human ear has good anti-noise performance and recognition ability in the process of speaker recognition.Therefore,more scholars and researchers are devoted to the study of the auditory characteristics of the human ear.Some algorithms based on the auditory characteristics of human ears have good recognition performance,but still have poor recognition performance under noise environment.To solve this problem,this paper studied human auditory characteristics algorithm and found that Cochlear Filter Cepstral Coefficient(CFCC)had a relatively good recognition effect in a noise environment.Therefore,CFCC is further studied in this paper to improve the overall recognition effect of its algorithm.The main work of the paper is as follows:1.A CFCC algorithm combining nonlinear power function and Wiener filter is proposed.First,the signal-to-noise ratio(SNR)of speech signal is improved by wiener filtering algorithm;Then,CFCC is fused with nonlinear power function to obtain a new CFCC,which improves the recognition rate of the algorithm under the condition of high SNR,and then the speech signal with improved SNR is taken as input speech signal of the new CFCC.Finally,the classifier is used for classification and recognition.The experimental results show that compared with the Mel Frequency Cepstrum Coefficient,Gammatone Frequency Cepstrum Coefficient and CFCC,the proposed algorithm improves the overall speaker recognition rate and has good robustness under noise environment.2.A new feature algorithm,MFCFCC,is proposed by integrating Mel filter bank into CFCC.First,the speech signal is input into the Mel filter bank for filtering and becomes a two-dimensional speech signal.Then,the two-dimensional speech signal is transformed into a one-dimensional speech signal by the weighted sum method,and the resulting speech signal goes through the CFCC process.Finally,the classifier is used for classification and recognition.The experimental results show that MFCFCC,compared with other algorithms,can significantly improve the speaker recognition rate in noisy environment and improve the recognition performance inpure speech.