Research The Text Analysis Key Technologies Of Uyghur Trainable Synthetic System Based On HMM

With the rapid development of information technology, the signal processing andelectronic science technology have a lot of progress. Speech recognition, speechsynthesis and natural language technologies get into a boom. At the sametime the Uighur speech recognition, speech synthesis and natural language researchwork also received a lot of results. For example, the development and implementationof Uyghur speech synthesis system base on a large corpus. But the overall systemperformance can hardly satisfy the actual needs. In the new information age, in orderto further improve the practical value of the Uyghur speech synthesis system, we needto greatly improve the degree of naturalness, clarity and, intelligibility. So, we havechosen the most typical trainable synthesis method that based on HMM module. Inthis thesis, the research work and results are as follows.Firstly, introduce the basic principles and algorithms of the trainable syntheticframework based on HMM module, and propose the Uyghur trainable syntheticsystem.Secondly, mainly research the front-end text analysis module. The key technologyresearches including special symbol processing, automatic labeling, unknown wordprocessing and prosodic phrase hierarchy and division. In order to improve thethe robustness of the text, we analyzed credibility of the text analysis module. Theback-end part preliminarily design out the context attributes and questions based onthe Uyghur language grammatical features, and made some optimization.Again, in this thesis we rebuilt the corpus, establish a large of rule base, the treebase and knowledge base, then made improvement.Finally, we made evaluation to the synthetic voice from a different perspective.The actual synthetic results further prove that establishment of the new system andnew methods of use are fruitful.