Research Of Noise-robust Technology In Conjunction Speech Recognition Based On HMM

With a half-century of development of speech recognition, our theoretical studies are very sophisticated. The speech recognition systems have achieved a high recognition rate in laboratory environment, and have gotten extensive applications. However, it is also far from the goal that the speech recognition systems have the same capacity with the people. Speech recognition still has many problems in the process of practicality such as voice environment, and so on. It also required a great deal of targeted researchs at all levels, which can make practical use of speech recognition.For increasing the voice recognition rate of Mandarin Chinese speech recognition system based on HMM model under additive noise environment, A comprehensive study of theoretical basis for speech recognition has been provided, and the anti-noise technology has been focused on. This paper adopts a progressive approach which builds an isolated-word speech recognition system, and then implement the conjunction speech recognition system based on isolated-word.A new endpoint detection algorithm has been presented based on complexity measure which has excellent anti-noise ability after researches on speech enhancement and endpoint detection. The results of tests show that the method can detect accurately in low signal to noise ratio. Anti-noise feature extraction technology has been in-depth studied in this paper. A new compensation method has been described for dynamic model. Experiments show that the characteristic parameter has good performance in robustness. The solutions to the problems in the implementation of HMM model such as the problem of many training set, the selection of initial HMM model, the data overflow and so on will be described in detail. It proved that the system has good robustness to noise.In this paper, the in-depth study and practice have been expanded on various parts of speech recognition, the exploratory works have been made, and have achieved a certain degree of success, which will laid a foundation for the continue research in this field.