基于共振峰过渡的协同发音语音合成算法.docVIP

第 43卷 第 9期 2010年 9月 天 津 大 学 学 报 Journal of Tianjin University V ol.43 No.9Sep. 2010 收稿日期:2009-07-03; 修回日期 :2009-11-30. 基金项目:国家自然科学基金资助项目 作者简介:康广玉(1975— ,男,博士研究生,讲师 . 通讯作者:康广玉, kgycat250318@. 基于共振峰过渡的协同发音语音合成算法 康广玉 1,2,郭世泽 3,孙圣和 1 (1. 哈尔滨工业大学自动化测试与控制系,哈尔滨 150001; 2. 天津工程师范学院自动化系,天津 300222; 3. 中国人民解放军总参 54所,北京 100001 摘 要 :协同发音是音素在语流中受前后音素的影响 , 而使 2个以上音素的发音特征同时出现的现象. 大语料库语音 拼接算法获得的语音合成效果最好, 但录制语料困难、 算法复杂、 耗费时间, 需要占用很大的存储空间, 在应用中存在困 难. 为此, 针对小语料库语音合成, 提出参数修改和波形拼接相结合的语音合成方法, 通过参数修改方法合成协同发音 段, 再与自然语音进行波形拼接. 实验表明, 该算法合成的语音效果较好, 而且对语料库内容要求较低. 关键词 :语音合成;协同发音;共振峰;基频轨迹 中图分类号 :TP391 文献标志码 :A 文章编号 :0493-2137(2010 09-0810-05 Speech Synthesis Using Formant Transition Based Co-Articulation KANG Guang-yu1,2, GUO Shi-ze3, SUN Sheng-he1 (1. Department of Automatic Test and Control, Harbin Institute of Technology, Harbin 150001, China ; 2. Department of Automation Engineering, Tianjin University of Technology and Education, Tianjin 300222, China ; 3. No.54 Institute from Headquarters of the General Staff, People’s Liberation Army, Beijing 100001, China Abstract :Co-articulation is the phenomenon of co-occurrence of two or more phonemes features due to the effect of front and rear phonemes during articulation. While the speech mosaicing algorithm for the large speech database ob-tains good performance on speech synthesis, it finds difficulty in practical application resulting from difficulty inthe collection of speech samples, the complexity of algorithm, the time consumed and large storing space. This paper proposes a speech synthesis scheme for small speech database based on parameter adjustment and wave mosaic of speech band. The co-articulation speech band was first synthesized with parameter modification, and then mosaiced with natural speech. Simulations demonstrate that the proposed algorithm performs well on speech synthesis and haslower demand on the size of speech database. Keywords :speech synthesis; co-articulation ; formant ; pitch trace 协同发音是指在发音时, 音素在语流中受前后音 素的