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

• 开发研究与工程应用 • 上一篇    

稳定音色的音乐语音变调方法

张栋,彭建云,余春艳   

  1. (福州大学 数学与计算机科学学院,福州 350108)
  • 收稿日期:2017-03-07 出版日期:2018-03-15 发布日期:2018-03-15
  • 作者简介:张栋(1981—),男,副教授、博士,主研方向为下一代网络、多媒体智能处理;彭建云,硕士研究生;余春艳,副教授。
  • 基金资助:
    福建省高校产学合作科技重大项目(2016H6010);福建省科技引导性重点项目(2016Y0060)。

Music Voice Modification Method for Stable Timbre

ZHANG Dong,PENG Jianyun,YU Chunyan   

  1. (College of Mathematics and Computer Science,Fuzhou University,Fuzhou 350108,China)
  • Received:2017-03-07 Online:2018-03-15 Published:2018-03-15

摘要: 传统变调方法由于未考虑音色和边缘音,会影响音质并造成失真。为此,提出一种改进线性预测的变调方法。根据音乐语音存在边缘音的特点,借助谐波冲激分离将其分为谐波信号和冲激信号,基于线性预测模型,将谐波信号分解为声道传输函数和声门脉冲激励信号,采用重采样对脉冲激励信号进行变调处理,利用帧信号叠加合成提高拼接段的连续性,通过频域的谐波冲激叠加合成重构音乐语音信号。实验结果表明,该方法能使音乐语音信号在变调处理后改变音高,保持音色相对稳定不失真,且语音质量比传统的线性预测变调方法有较大提高。

关键词: 语言信号, 歌声合成, 频谱包络, 音色, 线性预测

Abstract: Traditional methods of speech modulation fail to consider the tone and the edge tone,which affect the sound quality and cause distortion.In order to solve this problem,this paper proposes an improved modulation method based on linear prediction.According to the perceptual characteristics of the edge tone existed in the musical sound,it uses the harmonic-percussive separation method to split an input audio signal into harmonic component and percussive component.It separates the harmonic component into resonator filter and glottal excitation signal based on linear prediction model.It resamples the glottal excitation signal to perform pitch-scale modification,and adopts frame-signal superimposed synthesis to improve the continuity of the splicing segment.Finally,it reconstructs the signal of musical sound by the harmonic-percussive superposition in frequency domain.Experimental results show that,compared with the traditional linear prediction modulation method,the proposed method not only can modify the pitch of musical sound while keeping the tone relatively stable and undistorted after the modulation process,but also can greatly improve the quality of speech.

Key words: voice signal, singing voice synthesis, spectral envelope, timbre, linear prediction

中图分类号: