Research On The Mandarin Continuous Speech Recognition Of Medium Vocabulary

The most purpose of speech recognition is to make the communication between people and machines more convenient. Through its development in half a century, the speech recognition system has become more and more perfect and many products have already come into our daily life.Although the technologies of speech recognition have been developed so maturate, the recognition rate of speech in noisy environment is still low. When the system which was trained in quiet lab environment applied to noisy environment, the performance would like to drop badly. This is mainly because the speech recognition is affected by the noise greatly, and how to dwindle the noise's influence is difficult.The main domain of speech recognition related to, and the development history and current research situation at home and abroad are introduced in the paper. The basic principles of speech recognition including the classify ways of speech recognition systems, the selection of speech recognition units, the technology of the feature parameters extracting, the technology of training of model and template matching, etc. are specified. And then the problems of speech recognition facing at present are discussed.The process of speech recognition is studied. And the methods of extracting the different main feature parameters of speech recognition, especially the extraction of LPCC and MFCC parameters which reflect cepstrum characteristics and have good performances are discussed.The mainstream technology of speech recognition acoustic modeling should be the HMM which models the system based on statistic method. The primary principles of HMM and its applications in speech recognition is introduced in the paper. The localizations of HMM are discussed at last.Finally, several experiments are performed to test the steps and principles of speech recognition. In the experiments the performances of different feature parameters and different HMM states are displayed, and the effect of spectral subtraction speech enhancement algorithm is validated.