Research On Problems Of Text-To-Speech System

Text-to-speech system (TTS) is a popular research object in the realm of speech processing in recent years; it converts text into speech through computer and then output the speech. A successful TTS outputs speech with clear timbre and natural fluency.This paper focuses on the prosodic structure prediction model. Therefore, firstly, label corpus manually through studying philological prosodic structure; secondly, use the decision tree algorithm to train and evaluate the corpus, thus get the prosodic structure prediction model. Then join the model to the TTS so as to obtain a good result.The speech synthesis module in the TTS adopts TD-PSOLA algorithm, which requests original speech waveform and synchronous mark of pitch. To label prosodic mark to the speech database, firstly speech segment should be segmented, then the syllable segment, initial/final segment, pitch label can be finished. This paper proposed the algorithm of speech segmentation, the speech segment of speech satabase can be segmented.