Research On The Technology Of Automatic Segmentation For Text-To-Speech System

The TTS technology is an important branch in information processing and finds its great use in the artificial intelligence. The core of TTS technology mainly focuses on the text analysis and rhythm control of voice. The former is the basic, mainly including special symbol conversion and word segmentation. The accuracy of automatic segmentation has great influence on the natural degree of subsequent module output voice flow, which dorminates in the text analysis system.The main objective of the paper is to design and realize a Chinese word segmentation system. After analyzing the major difficulties appearing in the automatic segmention, the purpose is to reduce the difficulty and improve the accuracy. The segmention algorithm integrated upgrading dictionary with mechanical segmentation was adopted, and using forth-back matching segmentation to eliminate the ambiguity. Improvement has been made in the following two parts: the first one is segmentation dictionary, which divides the singular one into the basic one and the characteristic word one. In the process of matching, the integral characteristic dictionary improves mechanical segmen-tation greatly and correct segmentation ratios of names, place and quantifier, and at the same time, it reduces the ambiguity caused by these words, saves the time in processing ambiguity and accelerates the segmentation.The second is the improvement in mechanical segmentation, which realizes two-way matching of the forth and back segmentations, and can select the forth or back matching in the segmentating. Simultaneously, the system realizes the screen segmentation and file segementation. Comparing with the former singular one, the system provides two segmentation patterns, and through comparison of the segmentation results, the accuracy of segmentation gains much advantage. According to personal likes and utility, the screen segmentation and file segmentation can be choosed, which is at the benefit of users. The testing result shows that the speed and accuracy of segmentation algorithm is rather high, and the algorithm is fairly accurate in processing the ambiguity, which meets the practical requirements of Chinese analysis in TTS. However, there are still some shortcomings in the system such as in processing ambiguity, in solving the problems encountering in automatic segmentation, and in unknown words and the ambiguity.