北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2015, Vol. 38 ›› Issue (3): 28-33.doi: 10.13190/j.jbupt.2015.03.003

• 论文 • 上一篇    下一篇

基于语音截止频率的声码器激励模型

徐静云1,2, 赵晓群1, 李荣芸1, 王峤1   

  1. 1. 同济大学 电子与信息工程学院, 上海 201804;
    2. 湖州师范学院 工学院, 浙江 湖州 313000
  • 收稿日期:2014-11-08 出版日期:2015-06-28 发布日期:2015-06-26
  • 作者简介:徐静云(1980—), 男, 博士生, E-mail: 1210488@tongji.edu.cn; 赵晓群(1962—), 男, 教授, 博士生导师.
  • 基金资助:

    国家自然科学基金项目(61271248);湖州师范学院校级科研项目(2014XJKY47)

Vocoder Excitation Model Based on Voicing Cut-Off Frequency

XU Jing-yun1,2, ZHAO Xiao-qun1, LI Rong-yun1, WANG Qiao1   

  1. 1. School of Electronic and Information Engineering, Tongji University, Shanghai 201804, China;
    2. School of Engineering, Huzhou University, Zhejiang Huzhou 313000, China
  • Received:2014-11-08 Online:2015-06-28 Published:2015-06-26

摘要:

在低速率声码器中,对激励信号的描述直接影响重建语音的质量. 为了改善音质,提出一种基于语音截止频率的声码器激励模型. 该模型编码时通过语音截止频率将激励谱分成谐波和噪声2个子带,谐波子带的激励谱幅度引入离散余弦变换变维模型进行描述,语音截止频率进行4 bit非线性量化. 解码时将恢复出的谐波子带激励谱幅度进行傅里叶反变换,噪声子带则由白噪声进行以语音截止频率为阻带截止频率的高通滤波,最后由谐波子带和噪声子带叠加出激励. 实验结果表明,该模型提高了全带激励谱幅度和谐波噪声成分的描述精度,可使重建语音的音质得以明显改善,主客观指标更优,对男声更为突出.

关键词: 声码器, 谐波噪声激励模型, 语音截止频率, 离散余弦变换

Abstract:

A vocoder excitation model based on voicing cut-off frequency(VCO) was presented. In encoding part, the excitation spectrum was divided into two distinct spectral bands by VCO: harmonic sub-band and noise sub-band, the model of variable dimension through discrete cosine transform was used to express the excitation spectral parameter of harmonic sub-band, and VCO was quantized through 4bits nonlinear scalar quantization. In decoding part, the recovered excitation spectral parameter of harmonic sub-band was inversely Fourier transformed, the noise sub-band was obtained by the white noise pass through a high pass filter which used the VCO as the stop-band cut-off frequency, harmonic sub-band and noise sub-band were superimposed to get the excitation. The model greatly improves the description precision of the entire spectral envelope and harmonic plus noise components. With better subjective and objective indicators, especially for male's speech, the reconstructed speech shows more natural.

Key words: vocoder, harmonic plus noise excitation model, voicing cut-off frequency, discrete cosine transform

中图分类号: