中文HMM参数化语音合成系统构建  

HMM-based Mandarin Speech Synthesis System

作　　者：胡克[1] 康世胤[1] 郝军[1] 

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046

出　　处：《通信技术》2012年第8期101-103,108,共4页Communications Technology

摘　　要：在语音合成领域,大语料库拼接合成方式有一些固有弱点,例如语料库建设成本过高,合成稳定性差等。而基于隐马尔可夫模型(HMM)的语音合成技术在多样化语音合成、多语言支持、系统资源占用方面优势明显。分析了基于HMM的参数化语音合成技术的基本结构和核心算法,研究语料库建设,声学参数提取,建模单元和HMM拓扑结构选择等问题,给出适合于中文语音的参数设置,实现基于HMM的参数化中文语音合成。In the field of speech synthesis,the synthesis corpus stitching has its inherent flaws,such as high cost of corpus construction and poor synthesis stability.However,HMM-based speech synthesis is of obvious superiorities in flexibility,multilingual support and resources occupation.The HMM-based parametric synthesis technique is studied,and various configurations including speech unit,HMM topology,acoustic parameters and context information are carefully designed and implemented,and thus a HMM TTS system for mandarin synthesis is built up.

关 键 词：语音合成 隐马尔科夫模型 参数化系统 

分 类 号：TN912[电子电信—通信与信息系统]