Specific Isolated Word Chinese Recognition System

Along with the research of speech recogonition, the technology for mediurn/small-vocabulary, speaker-dependent and isolated-word speech recognition has come to the age of maturity, now, it is used in fields such as household electronics and intellectual toys and so on. This thesis focuses on the application and algorithm for medium/small-vocabulary, speaker-dependent and isolated-word speech recogonition. The results from this research will found the basis for speech recognition kernel in speech-controlled electronic devices.a process for speaker-dependent and isolated-word speech recognition system is presented starting from preprocessing i.e. digital sampling, noise filtering, windowing and so on. It is followed by endpoint detection and pick-up parameters of speech signal. Methods for speech recognition are discussed as well. Experimental results are analyzed and compared in detail. A major improvement in recognition accuracy has been achieved in DTW algorithm compared to classical methods, eventually a software for speaker-dependent and isolated-word speech recognition is worked out.These are the main contents of this paper:First, pretreatment and feature selection are discussed. It is followed by digital sampling, windowing, noise filtering, endpoint detection, time-domain vector and transform-domain vector. Double-threshold are adopted in the part of endpoint detection.Second, we talked algorithms of speaker-dependent and isolated-word speech recognition, including DTW, VQ, HMM, and ANN.Third, LPCC and MFCC are compared in detail, also recognitionaccuracy for total distance and average distance are talked over. A MFCC are used ultimately in feature selection, furthermore, recognition accuracy has been achieved greatly in DTW algorithm by widen the endpoint.Forth, a speaker-dependent and isolated-word speech recognition system is designed. At the same time, we made some correlative experiments for discussion.