作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• 多媒体技术及应用 • 上一篇    下一篇

基于谐波加噪声激励模型的改进语音合成算法

戈永侃,于凤芹   

  1. (江南大学 物联网工程学院,江苏 无锡 214122)
  • 收稿日期:2015-12-10 出版日期:2016-12-15 发布日期:2016-12-15
  • 作者简介:戈永侃(1991—),男,硕士研究生,主研方向为语音信号处理;于凤芹,教授、博士。

Improved Speech Synthesis Algorithm Based on Harmonic Plus Noise Excitation Model

GE Yongkan,YU Fengqin   

  1. (School of Internet of Things Engineering,Jiangnan University,Wuxi,Jiangsu 214122,China)
  • Received:2015-12-10 Online:2016-12-15 Published:2016-12-15

摘要: 传统基于隐马尔科夫模型(HMM)的语音合成算法使用高斯白噪声和脉冲串来表示清浊音的激励信号,合成的语音较为嘈杂。为提高合成音质,基于谐波加噪声激励模型,提出一种语音合成算法。将语音信号逆滤波得到声门波信号,对声门波信号进行谐波分析提取谐波成分,并计算谐波成分的线谱对参数作为谐波特征进行HMM训练。在语音合成时根据新生成的特征参数重构出低频段谐波部分与高频段噪声部分,并将两者混合作为语音的激励信号进行语音合成。实验结果表明,与基于脉冲激励的语音合成算法相比,该算法生成的语音频谱更接近自然语音,并且能够有效地减轻合成语音的机器声,提高合成语音的自然度。

关键词: 语音合成, 谐波加噪声模型, 激励信号, 逆滤波, 隐马尔科夫模型

Abstract: The excitation signal used in the traditional Hidden Markov Model(HMM)-based speech synthesis algorithm is either a pulse train or white gaussian noise,and the synthesis speech sounds buzzy.An improved speech synthesis algorithm based on harmonic plus noise excitation model is proposed to enhance the quality of speech.After inverse filtering,the harmonic signal in glottal flow is extracted and modeled by Linear Spectrum Pairs(LSP) coefficients.The LSP coefficients are sent into HMM training as the harmonic feature.In synthesis stage,the harmonic part and the noise part are reconstructed from the newly generated coefficients and mixed together as the excitation of speech signal.Experiment results demonstrate that the excitation generated by this algorithm is more accurate compared with speech synthesis algorithm based on pulsed excitation.The algorithm can effectively relieve machine noise of synthesized speech and improve the naturalness of the speech.

Key words: speech synthesis, harmonic plus noise model, excitation signal, inverse filtering, Hidden Markov Model(HMM)

中图分类号: