论述了采用基于统计模型进行韵律建模的思路。在此基础上,提出了基于音节韵律特征分类的韵律建模思路,并采用韵律模板和韵律代价函数实现了韵律的自动预测。对该模型的自动训练算法进行了详细的阐述。根据统计的韵律建模方法,还分析了韵律特征间相互关联对音节韵律模板选取的影响。最后,进一步分析了统计韵律模型的进行韵律预测的误差分布情况,表明了该模型能够使语音合成系统具有较高自然度和高灵活性的特性。 Discusses the idea of ​​using prosodic modeling based on statistical models. On this basis, the prosody modeling method based on syllable prosody classification is proposed, and the prosody prediction is realized by using prosodic template and prosodic cost function. The model of automatic training algorithm is described in detail. According to the statistical prosodic modeling method, the influence of interrelation of prosodic features on the selection of syllable prosodic templates is also analyzed. Finally, the error distribution of prosodic prediction in statistical prosody model is further analyzed, which shows that this model can make the speech synthesis system have high naturalness and flexibility.