The Research And Realization Of Uighur TTS System Using Variable Length Concatenating Units

Multi-lingual text-to-speech is an interesting research subject in TTS research field. To realize multi-lingual TTS system especially minority language include multi-dialect is also an interesting research subject of Chinese TTS research field. Due to the ethnic structure in the Uighur Autonomous Region of Xinjiang province and the importance of its physical location, it has significant social benefits and broad application prospect to study the text-to-speech system.In accordance with the main problems encountered in traditional TTS system, The article focused on the summary of prosodic rules of nature speech, design of speech corpus, synthesis methods and so on, a Uighur TTS system based on variable-length concatenating units is realized. Firstly, this article solves problems in the Uighur text analysis, such as syllable separation ,and summarize stress?￠pause and tone rules of prosodic based on the features of the Uighur language and phonetics. Then propose the design of ?°Context Vector?±, and uses Greedy algorithm to optimize corpus. At last, this article introduces synthesis method using variable-length concatenating units. Select the bigger unit firstly, when the synthesis of unit is syllable, introduce Viterbi algorithm to select the best unit synthesize speech based on context.This system can synthesize speech with quite high naturalness and maintain the speaking style of original speaker. So a new method in Uighur speech synthesis has been developed.