Research On Important Problems Of Isolated-Word Speech Recognition System

As the most comfortable communicating method between human beings, speech method is obviously one of the best ways in man-machine interface. Nowadays, the speech recognition systems are making our lives different and different. Accompanied with the development of the embedded system technology, the user devices are becoming smaller and the users require more simple and convenient operation. Therefore, the speech input method based on isolated-word speech recognition system is a very good choice for man-machine interface. All mentioned above will provide the isolated-word speech recognition a bright future.This paper introduced the history and actuality of the speech recognition technology, and the basic theories related with speech recognition, including speech signal processing, acoustics feature extracting, HMM and so on. Further more, a large-vocabulary, speaker-independent, dynamic-vocabulary isolated-word speech recognition system is built as the base-line system. The vocabulary is consisting of 600 words, including person-name, place-name, command word and daily word. Researches in some important problems of speech recognition system are proposed on this base-line system.This thesis researches speech point detecting arithmetic. A speech point detecting arithmetic based on short-time energy has been presented on the basis of analyzing and testing the common arithmetic of speech point detecting. A series of test has proved the arithmetic cost less computing resources, and provide a high accuracy, and perform well in noise environment. On the other hand, a voice active detecting arithmetic published in the distribute speech recognition system standard of the European Telecommunication Standardization Institute is also introduced in this paper. An automatic recording system has been constructed based on this arithmetic, which achieved the function of detecting the input speech to starting and ending the...