Study On Speech Recognition Technique Based On HHT

Speech recognition is a comprehensive subject, involving many scientific fields, such as physiology, psychology, phonetics, calculator and communication. The development of speech signal processing and recognition technique can not get away from that of other related subjects, but its ultimate purpose is to realize the communication between human and machine in a natural language. Speech recognition technology represents the important development direction of calculator technique. With the approaching of the multi-media ages, it becomes more and more urgent to solve the hard problems of automatic speech recognition.The speech signal is a kind of typical non-stationary signal. But in a long period of time, the speech analysis and expression uses stationary analyzing methods, based on the assumption that the signal is stationary in short time. Though these methods obtained the large success in actual application, it is still different to human's feeling ability. In recent years, the development of time-frequency analysis and wavelet transformation theory provides a new means for speech analysis, and some fields have been progressed obviously. The Hilbert-Huang transformation technique (HHT) absorbs the advantage of wavelet transformation 'multi- resolution, as well as overcomes the difficulty that il is needed to choose an appropriate wavelet function for wavelet transformation. This kind of dative decomposition method is suitable very much to analyze nonlinear and non-stationary process.This thesis applies HHT technique to endpoint detection, noise repression, pitch detection and the characteristic parameters procurement in speech Recognition field. It not only analyzes this means feasibility, but also verifies this idea by experiments and obtains satisfying results. HHT technique show us a new...