The Research On Tibetan Speech Recognition Technology

Speech recognition is a technique that computer changes the human voice signal into the corresponding text or command by identifying and understanding process. Speech recognition research began in the 1950 s. since Bell Labs successfully study the world’s first recognize ten alphanumeric pronounce experimental system in 1952, more and more state institutions and companies concerned the speech recognition technology, and invested heavily in research and development. Nowdays, speech recognition technology has become a hot spot of information processing research. Speech recognitions research on Chinese and English other languages have made great achievements, but speech recognition research on Tibetan and other ethnic minorities language in China lies in preliminary stage.This thesis elaborates the selected topic significance, the research value of the thesis and a general overview of speech recognition technology, introduces the structure of the Tibetan syllables and the characteristics of Tibetan language pronunciation, analyzes the characteristics of the modern Tibetan speech and selection of the Tibetan speech recognition unit, introduces in detail the establishment of Tibetan speech recognition corpus, the establishment based on Tibetan speech dictionary of sound vowel structure, the establishment of Tibetan speech recognition of statistical language model and the acoustic model training based on hidden markov model.Based on Acoustics, Speech and linguistics-based voice processing mechanisms,this thesis has studied speech recognition technology, and it firstly has studied the internationally relatively mature speech recognition technology. Combining the analysis of the knowledge of Tibetan language linguistics and Central Tibetan language phonetic features, this thesis has preliminary achieved Tibetan Central Tibetan language speech recognition technology, on the bese of which, it has developed Tibetan continuous speech recognition system based on HTK toolkit speaker-independent small vocabulary, and Tibetan speech recognition system based on the speech recognition chip LD3320 simulating the man-machine interaction.