The Study Of Prediction Methods On The Mongolian Prosody

In recent years, the speech synthesis technology is developing very quickly. It has brought great convenience to people’s life, but the naturalness of output speech is still not satisfied. The prosody prediction is one of the key steps of speech synthesis technology. So if you want to improve the naturalness of synthesized speech, the most important thing is to grasp the structure of rhythm and improve the ability of prosody prediction. At present, there are many researches on the Chinese prosody. However, the researches on the Mongolian prosody are relatively few. In the thesis, we studied the methods of the Mongolian prosody.In this thesis, we analyze the rhythm structure of Mongolian which is composed of rhythm and rhythm phrase. The prosody prediction methods mainly include two categories:one is based on rules and the other is based on statistical learning. In the experiments, we use three methods to deal with Mongolian prosody prediction, which are C5.0, ID3and CRFs. The implement of C5.0is provided by commercial software Clementine. The ID3algorithm is implemented by ourselves. And CRFs is achieved by CRF++. On the comparison of the three methods, the result of CRFs is the best, as it contains relatively comprehensive context information; both C5.0and ID3belong to decision tree algorithm, but C5.0use Boosting propulsion technology and gets a better effect than ID3.