Research On Isolated Word Speech Recognition Algorithm And Realization On DSP

Speech recognition is a part of the language chain, which is the study of the machine can accurately identify the voice content of the person. The ultimate goal of speech recognition is to make computers understand any person, any content of speech. With the increasing application requirements, the recognition rate and robustness of the traditional speech recognition algorithms are not satisfactory. This paper analyzes the basic theory and algorithms of speech recognition, we conducted in-depth study of the theory of feature extraction and template training, the simulation result shows that compares with the tradition algorithm, the new algorithm improves the recognition accuracy.The main jobs of this thesis are follows:(1) Introduces the basic concepts of speech recognition technology. The status of speech recognition technology are described and the difficulties of speech recognition are analyzed.(2) The generation of speech recognition and the speech recognition system are systematic analyzed. This paper researches the pre-emphasis, add window and voice detection of speech recognition, the processes are performed with MATLAB.(3) LPCC and MFCC are researched. Because of the deficiency of the tradition algorithm in noisy environment, this paper proposed an improved algorithm which can improves the recognition accuracy and the robustness for considering the noise corruption of each sub-band by adding a weight coefficient.(4) A technique for preparing reference templates are proposed, the developed technique improved way for preparing reference templates. This technique extracts the reference template from a set of examples by computing the average length, the simulation results show that compared with using a single reference template, this technique improves the recognition accuracy and the robustness. Besides, compared with the VQ, it needs fewer examples and reduces the complexity of the algorithm. The experiments get fine identification effect.Considering the noise influence, an improved spectral subtraction method is researched and analyzed. The simulation results show that the algorithm reduces the noise effect.(5) Speech recognition algorithm is implemented by the TMS320VC5416 DSP. Algorithm code was compiled with C language in integrated development environment (CCS), the algorithm code was implemented by simulating and debugging. The simulation result proves the effectiveness of the speech recognition algorithm implementation by DSP.