基于HMM和Monte-Carlo方法的生成基频曲线的新方法

A New Method Generating Pitch Contour Based on HMM and Monte-Carlo Method

提出了一种基于隐马尔科夫模型(HMM)的汉语韵律短语的统计基频模型,并尝试使用蒙特卡罗方法(Monte-Carlo)从模型生成基频曲线。该模型使用HMM作为基本框架,具有HMM理论体系所能支配的各种优点。并将韵律词作为模型单元,使得模型能够反映韵律层次级的连续变调。Monte-Carlo方法能生成相对模型最优的基频曲线,并保证生成的基频曲线具有足够的精确度。该文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 proposes a novel statistical Mandarin-prosodic-word pitch model based on Hidden Markov Model(HMM).And writers try to generate the pitch contour from the model with Monte-Carlo method.Based on HMM,our method takes all the advantages of HMM theory.In the meanwhile,to reflect the tone-changing phenomenon in continu-ous speech,this paper uses prosodic word as model unit.By utilizing Monte-Carlo method,the paper can not only gener-ate the best pitch contour corresponding the model but also assure its precision.Lastly,it presents the results of experi-ments on the model and future expansion is explored.