This thesis presents a strategy and related techniques for speaker recognition in noisy environment. Speaker recognition is one of important research branches of speech signal processing field, which has found many applications .There are six chapters in the thesis, presenting the topics: the study background, the basis of speech signal processing, speech signal enhancement, the endpoint detection of speech signal processing, speaker recognition and the characteristic parameter extraction, speaker recognition based on Hidden Markov Model, respectively.The main achievements of the thesis are as follows:(1)Extract out MFCC and one order difference MFCC characteristic parameters.(2)With the principle of spectral subtraction, through statistic analysis on noise, we realize speech enhancement of input speech signal, enhance the articulation and intelligibility of speech, approximately recover the primitive speech.(3) Through analysis on noise characteristic, we design a method to recognize background noise automatically, and we propose a speech endpoint detection algorithm based on subband subtraction. Adaptively distinguishes quasi quiet environment from noisy environment. And automatically detect word or phase in the sentences.(4)Design a data structure of HMM for the recognition system, realize recognition with Left-to-Right and Ergodic HMM models. Test two models with same data. To each model, we use different length train and recognition data. Test the shortest length of train and recognition data with perfect recognition effect...
|