基于HMM混合激励英语文语转换系统.pdfVIP

第 9卷第 6期 江南大学 学 报( 自然 科学 版) Vol. 9 N o. 6 2010年 12月 Journal ofJiangnanUniversity(NaturalScienceEdition) D ec. 2010 基于 HMM 的混合激励英语文语转换系统 雷冠军 (永城职业学院, 河南永城 476600) : 采用传统激励模型的基于 MM的文语转换系统的合成语音自然 不高, 主要原因是其采 用的激励模型用脉冲串作为浊音段激励和白噪声作为清音段激励过于粗糙, 不能很好地模拟自然 语音的激励借鉴MELP语音编码算法,把混合激励模型应用到系统中, 对合成语音激励信号进行 改进试验结果表明, 混合激励模型大大提高了系统的自然 : 隐马尔可夫模型; 语音合成; 文语转换; 混合激励 : TP 391 : A : 1671- 7147( 2010) 06- 0658- 04 HMMBasedM ixed Excitation English TexttoSpeech System LEIGuanjun (Y ongcheng V ocationalCollege, Yongcheng 476600, China) bstract:TheTexttoSpeech system based on MM which uses the traditional excitationmodelhas a low degree of naturalness, This problem occurs ow ing to the rough excitation scheme where the sw itch between pulse train and random noise are employed tomodel voiced and unvoiced segments respectively, and the rough excitation can notmimic the excitation of natural speech. Based on the MELP speech coding algorithm, the paper applies themixed excitation model to system to mi prove the excitation. The experiments show that the m ixed excitation greatly enhances the degree of naturalness to theTexttospeech system. Key words: hiddenmarkovmode,l speech synthesis, text to speech, m ixed excitation : , [ 1] , (, ), , , , ,, , MELP ,