Studies On Techniques For Corpus-Based Text To Speech System Based On CART

Speech synthesis technology is a leading edge in today's hot topics. This article discusses on pre-selection in corpus-based speech synthesis and the CART decision tree theory is applied in the process. This article proposes the method of based on the binary bits to express the pre-selection rule and has realized the pre-selection tree with the method. At last the article uses the pre-selection tree to realize a speech synthesis system with the nature and fluent speech.Speech synthesis based on a large corpus will be the most important way to get natural synthesis speech for a long period of time. The pre-selection of the corpus is one of the most important topic in a corpus based speech synthesis system and this article applies the CART decision tree algorithm in pre-selection, and design experiment to check its effect.This article proposed data expression method based on the binary bits and applies it in the expression of the pre-selection tree's rule. In this way, set and its subset can be expressed conveniently. The decision tree algorithm divides the data unceasingly with its rule until found corresponding classification. Through the utilization based on the binary bits expression method, determination may use a series of logical bit operation to realize, far quicker than the general match category algorithm.In view of the question that used the simple question set to the multi-dimensional data class possibly conduct tha data to excessive piece, this article has used the compound question set to optimized the method. The paper has carried on analytic statistics processing to the corpus data, designed a software TTS TRAIN to realize the pre-selection tree's foundation specifically according to the minimum cost-complexity principle to carry on the pruning.This article has given the speech synthesis system's structure and the module constitution in realizing the speech synthesis system based on the CART pre-selection trees. To fast the procedure of from the Chinese character to the Pinyin and the pre-selection tree, this article has designed the Chinese character zone code, the Pinyin and the pre-selection tree's index comparison document. The paper has applied magnanimous participle software to division grammar word level, used the artificial labeling method to train the C4.5 decision tree, realized rhythm level labeling with the C4.5 algorithm, carried on the pre-selection process with the pre-selection tree, and carried on the final choice using the Viterbi algorithm to. At last the article through the designed experiment this article has realized obtains the speech synthesis system's speech synthesis quality, which is between acceptable and good.