Study On The Text To Speech

Along with the development of computer technology, the pronunciation interaction has already become the essential method for man-machine interaction. Making the computer send out the natural and smooth pronunciation is the matter which people long for even in dreams. With the development of the linguistics, phonetics and computer technology, this goal is becoming more and more near. In recent years some products about the Text to Speech are gradually published, but can't be satisfying.The thesis mainly researched the contents of text analysis. It is the first module for text-to-speech conversion, for the purpose to make a good foundation for TTS (text-to-speech system). The text analysis approximately includes five parts: the documents structure analysis, the text normalization, the grammar analysis, the rhythm modeling and the Grapheme Phoneme Conversion. In this paper, the main results are:This thesis studied the commonly used word segmentation lexicon mechanism. On the basis, this thesis designed and structured the index table according to the the second word'order in Chinese table' colum. This index table may apply to the entire word segmentation lexicon , also may apply to the word by word segmentation lexicon. Experiments show the method can be used to improve the searching speed. The program laid the foundation for enhancing the speed of dividing words.After introducing several solutions to deal with words ambiguity, taking into account the results from using the combination methods are the segmentation of a sentence. Thus, this thesis processed the ambiguous words with N Grammar. Furthermore, in order to obtain the frequency of each word in N Grammar, this thesis used the Internet search engine to access word frequency statistics. These research activities laid the foundation for enhancing the speed of diving words, and provided a reference method for resolving the ambiguity, and had some reference value.Under the researching and contrast of the method of word segmentation, this thesis proposed two methods against the solution to overlapping ambiguities: One is the method of non-overlapping. After dividing the word with the method of word by word MM, overlap these words and come to the segmentation methods. The algorithm of non-overlapping method is simple and effective, but the time and space performance is poorer than than the dynamic programming methods. The other is based on the combination of segmentation method. According to the combination characteristics of the overlapping terms, firstly find out all the combinations, then eliminate all the combinations which are not conforming to the condition ,and obtain the possible segmentation methods of the words. This method is simple, and in the space-time performance is period than the existing segmentation methods. However, this method is addressing the issue of cross-ambiguous problems, not to the solution of other ambiguous problems.