Based on the work of text processing and prosody generation, the present research was conducted to explore the automatic mapping from Chinese sentence text to its corresponding prosodic features. According to the results of text processing and the sum-up of prosodic rules, a model of prosody generation was developed for the text-to-speech system in Mandarin Chinese.
First of all, a comprehensive dictionary as well as a special one and a regular one was designed for text processing. Second, lexical parsing and POS (abbr. part of speech) labeling for sentence text were accomplished with maximal matching
algorithm. Third, chunks of sentences were acquired by means of shallow parsing. In the end, sentence text was standardized and labeled with pinyins.
Rules of the setting of prosodic boundaries, pause and stress were integrated for prosody generation on the basis of the previous research. The prosodic model of sentence text was elaborated. Prosodic grades setting and prosodic parameters
generation were automatically realized based on the chunks of sentences and their
syntactical features.
The results showed that the leakage and error of unknown words were both reduced by hierarchical parsing in text processing and that the simulating of prosodic hierarchy was quickened through the combination of prosodic words. In addition,random factors for dynamically simulating prosodic features were introduced in prosody generation, and prosody rules stated qualitatively were quantitated. These were considered to be helpful for improving the spontaneity of synthetic speech. In this research, the results of basic study were applied to engineering technology. Although there were some problems remained to be resolved, it was hoped to conduce to the development of speech synthesis technology in China.
修改评论