Anti-noise Speech Recognition Technology Research Based On The ARM-Linux

Speech recognition has received more and more attention recently due to the important theoretical meaning and practical value. Up to now, most speech recognition is based on conventional linear system theory, such as Hidden Markov Model (HMM) and Dynamic Time Warping(DTW). With the deep study of speech recognition, nonlinear system theory method must be introduced to it. Recently, with the development of nonlinear-system theories such as artificial neural networks (ANN), chaos and fractal, it is possible to apply these theories to speech recognition. Therefore, the research in this recognition paper is oriented on the theory and application of the mixed model HMM-ANN, and the related algorithms and model are developed.Mandarin speech recognition technology and implement approach is studied in this thesis. Introduce the basis theory of speech recognition, include the math model of speech signal, pretreatment, CASA (Computational Auditory Scene Analysis) algorithms and feature parameter extraction algorithms, Aim at the occasion where the design was applied, improved the configuration of CASA, optimized its arithmetic, debate the mandarin speech recognition technology and the fundamental of CASA. The application of CASA distilled multiple pure speech signals. The speech feature extraction algorithms was expatiated and improved.The paper indicated the merit and defect of Liner Prediction Cepstrum Coefficient and Mel Frequency Cepstrum Coefficient. Discussed the extract method and operate process of MFCC detailedly. The performance of speech recognition and application characteristic of HMM (Hidden Markov Model) and SONN (Self Organizing Neural Networks) methods used in this paper is compared. Discussed the theory and model parameters of HMM, analyzed the extract method of each parameter and resolved three basic problems, explained the basic conception of ANN, BP network and SONN. The research improvement in this paper is oriented on the theory and application of the mixed model HMM-ANN, which is formed by the combination of the Continues Hidden Markov Mode (CDHMM) and SONN, and the related algorithms and model are developed. After the high-point list of speech signal was computed by means of CDHMM, For the same state, build the same dimension's speech character vector by DTW, and affiliate it into SONN speech recognition sort. The HMM-ANN model has the ability of modeling and static state classify.The paper designed the software and hardware structure of speech recognition system, researched the CASA and HMM-ANN model arithmetic under the ARM-Linux crossed complier, Tested speech recognition rate in several occasion. As result, compared by the former HMM model method, the ameliorative CASA and HMM-ANN arithmetic improved the veracity, anti-noise ability, the stability and self-adaptability of speech recognition, indicated model performance, proved the feasibility and validity, and indicate the direction of the research improvement.