Research On Text Analysis Of Text-To-Speech System Based Bayesian Network

Bayesian networks is one of the important methods that investigate the undecidable problem. It is based on the theory of probability and statistics and provided with the solid mathematical foundation. Due to the natural expression, strong reasoning ability, facilitate customization and many other advantages, Bayesian networks has been widely used in many areas. Text-to-Speech (TTS) is the technology that letter information is converted into sound signal and out according to the rules of dealing with sound, which will enable the computer to read fluently letter information, so that people can understand the content of the information by listening. The text analysis has greater uncertainty, because Chinese language is profound, so the traditional rule-based text analysis method can not be good to TTS system, in particular the open letter set system. In this paper, we use the theoretical framework of Bayesian network to the text analysis in TTS. This paper discusses the based knowledge of Bayesian network, the main problem in the text analysis and text analysis algorithm based Bayesian network , experimental results analysis.First of all, we discuss based knowledge of Bayesian network including the definition of that, symbol expression, and on the basis of which we discuss several typical Bayesian network classifier, for instance, Naive Bayesian Classification,Tree Augmented Naive Bayesian Classification,BAN Classification,Bayesian Multi-net Classification,General Bayesian Network Classification and so on, further their characteristics; In order to reduce the computational complexity of Bayesian network, we investigate context independent relationship , causal independent relationship and discuss the role of the independent relationship.Being directed at the difficulty in Text analysis ,we focus on the method of determination the sentence border, on the basis of that symbol as the main basis for the text analysis ,we give a decidable algorithm of misunderstanding symbol, and the examples of its application; Targeting special symbol there is method to analyses them ,and the English and figures in Chinese are classified and corresponding analyzing algorithm is given; We discuss the two main methods that rules-based approach and statistic-based approach ,moreover compare them; classification method of Multi-tone words is showed and according to the corresponding relationship between part of speech and phonologization of Multi-tone words , which be divided into class A and class B, and corresponding analyzing methods is given in view of the different types of Multi-tone words, further we discuss the analysis results.In this paper ,the analyzing word has two main methods: One is the rule-based method, and the other method is based on probability. the rules-based approach is an important means dealing with decidable problem. It has some advantages for instance sufficiently learning from experience of experts ,more intuitions and obtaining easier. the method based on the probability is provided with more robust and higher performance. As data analysis and decision support tools based on the theory of probability and statistics, Bayesian network is suitable for text analysis in Chinese which is profound. In this paper, text combine the text analysis and Bayesian network integration, and select the test data for testing. The test results show that the algorithm introduced in this paper is better than the others in accuracy rate of disparting word , R-measure, F-measure, and the analysis of the polyphonic word also show good performance in different the test set.