浊声基频轮廓对汉语合成自然度提高的分析与综合  

Analysis and synthesis of continuous voice pitch contour for improving Chinese synthetic speech naturalness

作　　者：田岚[1] 陆小珊[1] 杨霓清[1] 

机构地区：[1]山东大学信息科学与工程学院,山东济南250100

出　　处：《山东大学学报（工学版）》2003年第4期413-416,共4页Journal of Shandong University（Engineering Science）

摘　　要：连续语音浊声基频轮廓是影响合成语音自然度和表现力的一个重要因素 .本文采用序位调值分类统计法 ,对汉语连续语音音调动态特性作了系统分析 ,提出一种用于分析和分层产生汉语连续语音基频参数的数学模型 .模型充分考虑了汉语发音特点 ,归纳了语言表达中音调变化的各种可能 ,并相应设置了控制调整参量 ,相对完整而实用地表示了语言知识和基频参数之间的对应关系 .对一些典型自然语句进行了仿真实验 ,结果表明 ,该模型控制产生的合成基频轮廓和测试目标可达到满意的吻合 ,对有效改善TTS系统语音合成自然度作用明显 .The continuous speech Fo contour plays key role for the naturalness and emotion in text to speech conversion system. Based on statistics method and clustering at the sequence location of each syllable, we systematically analyzed a large number of Chinese continuous speech pitch contours. As a consequence, a hierarchical prosody analysis and synthesis model is introduced, in which Mandarin characteristics are fully taken into account, introducing all tone patterns and phrase dynamic trend, and setting relative control command parameters and sandhi rules. The model quantitatively describes the relationship between prosody features and Chinese multi layer linguistic information. The emulating tests for some typical natural utterances show that synthetic Fo contours have good correspondences with the objective samples and that the model is expected to improve the naturalness of TTS synthetic speech evidently.

关 键 词：文语转换 TEXT-TO-SPEECH 韵律特征 基频 语音自然度 浊声基频轮廓 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TN912[自动化与计算机技术—计算机科学与技术]