The Algorithm Research Of Speech Recognition Based On VQ And DTW

A period of the research of Speech Recognition set during the 1950s, through the work of technique research for 70 years, we broke through in many key technologies. In recent years, the high recognition rate machine of Speech Recognition is based on personal computer, and the algorithm is mainly implemented by writing software, so the system is very large. In order to contract a physical volume of the system, and also reduce the cost, most of all, to meet with practical applications, the next generation Speech Recognition chip will be hardware and software co-designed based on the use of special DSP, so we must have efficiently of arithmetic with mutually and permit to DSP.In this paper, a real time system of speaker recognition based on DSP has been designed. We improve the algorithm while design electric circuit, so the improving algorithm can be tested. All my works are as follows:Firstly, it is researched the theories and the universal frame of speech recognition, discussed the key technologies of speech recognition.Then, it is mainly elaborate the theories of VQ(Vector Quantization) and DTW(Dynamic Time Warping). And also tell of the 3 phrases for the complement of speech recognition, they are primary processing phrase,character distracting phrase and recognizing phrase.And then according to the characteristics of word-isolated speech recognition, bring forward new methods and improved algorithm for speech recognition, as in the phase of preliminary speech signal processing, combined with the two new methods to complete the pure speech paragraph selecting, the new methods are a new self-reliant and dynamic accommodating length of the window for the speech signal selecting algorithm,and a zero-energy-differences bounds measuring algorithm. As bring forward an improved VQ and DTW to the facility of DSP processing for speech recognition.Last, it is discussed the principle algorithm processed by DSP, and also brings forwards the flow chart of algorithm based on DSP processing; using the DSP as the central equipment to expand outer modules, speech signal collection circuit to complete the experimental system of speech recognition.The experiment shows that the arithmetic can meet with the DSP processing, and also guarantee high recognition rate, it have good recognition effect.