基于两层隐马尔可夫模型的可视语音合成  

Visual Speech Synthesis Based on Two-level HMM

作　　者：尹宝才[1] 李敬华[1] 贾熹滨[1] 孙艳丰[1] 

机构地区：[1]北京工业大学多媒体与智能软件技术北京市重点实验室,北京100022

出　　处：《北京工业大学学报》2006年第5期416-418,共3页Journal of Beijing University of Technology

基　　金：国家自然科学基金面上项目(60375007);北京市自然科学基金项目(4061001)

摘　　要：为了解决可视语音合成中语音与口型多对多的对应关系,本文提出基于两层隐马尔可夫模型的可视语音合成,该模型有效结合了语音和口型的上下文相关性,解决了语音与口型多对多的对应问题,合成出了准确率高、连贯、自然的口型序列,该方法具有完全自动化的特点．In order to solve the problem of many to many mapping between speech and lip in visual speech synthesis, this paper proposed a method based on two-level HMM for visual speech synthesis, which combined the context of audio and mouth effectively, solved many to many correspondence between audio and mouth, synthesized accurate, coherent and natural mouth sequence. And this way is also roboticized completely.

关 键 词：语音合成 映射 隐马尔可夫模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]