Abstract: A text analysis method for converting grapheme to dialect phoneme is proposed for statistical parametric dialect speech synthesis.A set of Speech Assessment Methods Phonetic Alphabet(SAMPA)-based symbols are designed for labeling pronunciation of dialect by comparing the differences between Mandarin and dialect.A set of conversion rules is also designed that can transform Mandarin pronunciation to dialect pronunciation.The text analysis is conducted for Chinese sentences to obtain lexicon words and their initials and finals.A transformation-based error driven learning algorithm is used to obtain the prosodic words and prosodic phrases boundaries.The conversion rules are employed to obtain the SAMPA of dialect initials and dialect finals.The input sentences are converted into context-dependent labels.Test result shows that the proposed method can generate correct context-dependent labels.

Key words: text analysis, grapheme-to-phoneme conversion, Speech Assessment Methods Phonetic Alphabet(SAMPA), speech synthesis, syntactic analysis

摘要： 为实现方言的统计参数语音合成,提出一种从文字到方言读音的文本分析方法。通过对比普通话和方言在声韵母方面的发音异同,设计方言的语音评估方法音标字母(SAMPA),用来标注方言声韵母的读音,得到从普通话读音到方言读音的转换规则。对输入的汉语文本进行分析,获得语法词、声母、韵母信息,使用基于转换的错误驱动学习算法获得语句的韵律词和韵律短语边界,利用普通话读音到方言读音的转换规则,获得方言发音的SAMPA音标,从而将输入的文本转换为统计参数语音合成所需的上下文相关标注。测试结果表明,该方法能较为准确地生成上下文相关标注。

关键词: 文本分析, 字音转换, 语音评估方法音标字母, 语音合成, 语法分析

CLC Number: 

• TP391