Research On Speaker Parameters Extraction Algorithm In Noisy Environment

Speaker recognition technology belongs to the category of biometric identification technology and is one of the most popular biometric identification technologies.With the progress of the Internet and the rise of artificial intelligence,speaker recognition technology has been widely used in various fields of real life.The extraction of feature parameters in speaker recognition technology and the establishment of recognition models are two key parts of the entire speaker recognition system.The extraction of speaker speech feature parameters is the most critical part of the entire system,directly affecting the entire speaker recognition system and the quality of performance is also a hot issue in current research.Since many current applications are based on an ideal environment,some common speaker feature parameters can achieve relatively high recognition rates.However,in practical applications,obtaining the scene around the speech is performed in a complex and non-ideal environment.At this time,using commonly used feature parameter extraction algorithms will lead to the performance of the speaker recognition system obviously drop under such a non-ideal environment.Therefore,this paper focuses on the speaker feature parameter extraction algorithm in the noisy environment.Based on the analysis of the current mainstream feature parameters,This paper proposes several improved algorithms of speaker feature parameters,and system is verfied under the simulation experiment environment and the actual environment in the Android mobile client.Experimental results show that the new feature parameter extraction algorithm proposed in this paper has better performance and better robustness than the traditional feature parameter extraction method.The main content of this paper is as follows:(1)The implementation process of a Hamming self-convolution window is introduced in detail and the Hamming self-convolution window is used in the process of speech preprocessing to reduce spectral energy leakage,which is beneficial to the extraction of speaker features.(2)Summarize several manifestations of present speaker's characteristic parameters,and introduce the extraction algorithm of Mel-Down feature parameters based on Mel filter banks and the extraction of auditory characteristic parameters based on Gammatone filter banks.(3)An improved algorithm based on Gammatone filter bank is proposed to extract the feature parameters and the two combined feature parameters are implemented.The simulation results of two experimens show that in the noisy environment,this paper proposes improved feature parameters and two combined feature parameters with a good recognition performance and robustness in speaker recognition system.In the end,the paper analyzes and summarizes to the improved speaker feature parameter algorithm proposed in this paper,and points out the deficiencies in the paper.In addition,in view of the problems existing in this paper,the method for improvment and research directions of future research are pointed out.