关键词: 文本分析, 语音合成, 上下文相关标注, 韵律预测, 字音转换

Abstract: This paper designs a six-level context-dependent label format, which includes an initial and final level, a syllable level, a word level, a prosodic word level, a prosody phrase level and a sentence level, for Chinese statistical parametric speech synthesis. The input Chinese sentence is firstly normalized and performs grammar analysis to obtain sentence structure and word segmentation information. Then the initial, final and tone of Chinese character are obtained by grapheme-to-phoneme conversion. The Transformation-Based error driven Learning（TBL） algorithm is finally employed to predict the prosodic word boundary and prosodic phrase boundary of the input sentence. Context-dependent labels of each sentence for statistical parametric speech synthesis are generated according to the context information obtained from above text analysis and prosodic prediction procedures. A Hidden Markov Model（HMM） based Mandarin statistical parametric speech synthesis is designed to evaluate the influences of different labels on quality of synthesized speech. Tests show that more context-dependent label information can achieve higher quality of synthesized speech.

Key words: text analysis, speech synthesis, context-dependent label, prosodic prediction, grapheme-to-phoneme conversion