【作者】 郭威彤； 梁青青； 裴东； 李锦珑； 杨鸿武；

【Author】 Guo Weitong, Liang Qingqing, Pei Dong, Li Jinlong,Yang Hongwu College of Physics and Electronic Engineering, Northwest Normal University, Lanzhou 730070, China

【机构】 西北师范大学物理与电子工程学院；

【摘要】 方言语音的转换是人机交互领域的一个重要研究课题。为实现方言语音的转换,需要建立方言语音的韵律模型。本文分析研究了西安方言的特点,设计了一个包括文本语料和语音语料的面向言语工程的西安方言语料库,文本语料由单音节字、双音节词、负载句和不同语调的语句组成;语音语料包括4个说话人的西安方言和普通话平行的语料。本文提出了基于归一化线性多项式的方言韵律转换模型与基于线性变换的方言时长转换模型。通过修改普通话的基频曲线实现普通话到西安方言单音节字和双音节词的转换。利用TBL算法划分韵律词,进而实现方言语句的转换。实验结果表明,转换出的方言语音与源方言语音具有很高的相似性。更多还原

【Abstract】 The conversion of Dialect speech is an important research topic in the field of human-computer speech communication. To achieve the conversed dialect speech, the prosodic model should be produced by the Text-to-Speech Synthesis system. The paper analyzed the acoustic features of xi’an-dialect, a Xi’an dialect corpus, which includes a text corpus and a speech corpus, was built for speech engineering. Text corpus consists of single syllables, two-syllable words, carrier sentences and sentences with different intonation. Speech corpus was recorded by four local speakers with contrastive (xi’an dialect and Mandarin) recordings. We proposed a dialect prosody conversion model based on the normalized liner polynomial method and a liner duration conversion model. Based on these models, we can get the single syllables, two-syllable words conversion from Mandarin. Dialect sentences can be achieved by TBL method. The evaluation results demonstrated the high similarity between the original xi’an speech and the converted speech from Mandarin.更多还原