Research On Noise Reduction Of Speech Technology Based On Sub-banding Andapplication On Speech Recognition

For the past few years, with the ceaseless progress of communication technology,the communication of audio signal become an international hot research topic. Inaudio communication, all kinds of interference noise is inevitable that some powerfulnoise signal may cause the performance of a sharp decline in voice processing system,and even make the system is in a state of paralysis. The research of noise reduction ofspeech rapid development in the past30years and become one of the focus areas ofthe current speech signal processing. Based on the analysis and studies of existingnoise reduction of speech technology, and reference to the document [31] and [42]handle with a method of sub-band adaptive filtering to noise reduction of speech, andcomplete simulation of algorithm and then applications into speech recognition, andfinally we obtained the experimental data and analysis of the experimental data bycomparing the experimental.The principal research of this article is organized as follows.(1) Based on the analysis of the present situation of study on noise reduction ofspeech, the existing technologies on noise reduction of speech classified. Thecharacteristics of speech noise and the speech noise to the influence on the speechrecognition are researched and analyzed, and it introduces the key techniques inspeech recognition.(2) The principles and elements of the sub-band filter banks are researched andanalyzed, the focus of research and analysis are the conditions of perfectreconstruction filter bank, and the four common sub-band filter banks are simulated.The four common sub-band filter banks were compared on the basis of a briefintroduction to speech quality evaluation methods, afterwards this article selects theWavelet filter bank as sub-band adaptive filter with partial decomposition filter bank.(3) On the basis of analysis of the adaptive filter, the focus of research andanalysis are the performance of LMS, including convergence, convergence speed,steady-state error and calculation, and we compare to the performance of LMS andnormalized LMS by using the simulation of MATLAB, design and complete thesub-band adaptive filter realization of this article, as well as compare the normalizedLMS with sub-band adaptive filter on the hand of convergence performance and calculation complexity, then it gets a result that the sub-band adaptive filter is fasteron the convergence speed and lower on the computational complexity.(4) This article completes the application of the sub-band adaptive filter on thespeech recognition. Firstly, the collection of all speech signals is completed andresearched. Secondly, the simple speech recognition system is completed by analyzingthe forward method of the speech recognition, as well as it improved the method onthe ordinary template matching (DTW algorithm) which we usually use in the speechrecognition, Compared this algorithm with the tradition algorithm, reducing the sizeof area of matching, The experimental prove that this algorithm in doesn’t influencethe recognition rate of speech recognition system, can enhance the system ofconsiderable recognition speed. Finally, the sub-band adaptive filter applied to thespeech recognition system, By comparison show that the sub-band adaptive filter hasa better performance.