基于隐马尔科夫模型的汉语韵律词基频模型

Putonghua prosodic word pitch model based on HMM

提出了一种基于隐马尔科夫模型（HMM）的汉语韵律词的统计基频模型。模型能反映韵律环境和基频曲线参数之间的映射关系,从模型可以估计一段基频曲线和一段文本之间的相关度,也可以从文本生成相应的基频曲线。本方法使用HMM作为基木框架,具有HMM理论体系所能支配的各种优点。同时将韵律作为模型单元,使得模型能够反映韵律层次级的连续变调。最后给出了实验结果并对模型的应用前景进行了展望。

This article proposed a novel statistical Putonghua-prosodic-word pitch model based on Hidden Markov Model (HMM). It reflects the relationship between prosodic environment and pitch parameters. With the model not only can we estimate the relationship between a section of text and a section of pitch contour, but generate the corresponding pitch from the text. Based on HMM, our method takes all the advantages of HMM theory. At the same time, to reflect the tone-changing phenomenon in continuous speech, we use prosodic word as model unit. At last, we present the results of experiments on our models and future expansion is explored.