Lhasa Voice Dialing Technology Research Based On Hidden Markov Model

The voice recognition technology is to make the computer or the phone through identification and understanding of the speech signal, turning it into the corresponding text or command technology.The development and application of voice recognition technology in the mobile phone will further extend this technology to the development of anew direction. Voice dial is a large application of this technology. At present, the domestic Chinese-based voice dialing technology has reached a very high level, The research on ethnic minority language the Tibetan Lhasa words voice dialing system are few and far between. Tibetan is Tibetan branch of the case of Tibeto-Burman of Sino-Tibetan. The Tibetan language is divided into the Wei Tibetan, Kang and Anduo three dialects in China. The differences between the dialects are mainly manifested in the voice. The Lhasa dialect is dialect Wei Tibetan. The Tibetan language belongs to the phonetic type language, Master spelling rules, spelling the word out. The Lhasa dialect in prosodic features co articulation own unique rules.In combination with the characteristics of the Lhasa dialect, Articles establish60pronunciations Lhasa discourse continuous speech corpus, in which30men and30women. Articles also research the Lhasa discourse continuous voice recognition primitive selection. Through acoustic analysis Lhasa words, articles select tri-phone as the basic recognition unit. Using international the mature voice recognition tools-Hidden Markov Model Tool (HTK), Articles establish Lhasa dialect tri-phone acoustic models. And use MFCC feature vector, decision tree, binding tri-phone increasing Gaussian mixture degree method to improve the recognition rate of the model. In a DOS environment, Article conduct recognition experiments using for test statement HTK tool. Finally, the male pronunciation voice dialing recognition sentence of recognition rate reached87.05%, word recognition rate of87.14%, the female pronunciation voice dialing recognition sentence recognition rate of86.45%, word recognition rate of87.31%。...