Journal of Computer Applications ›› 2016, Vol. 36 ›› Issue (12): 3374-3377.DOI: 10.11772/j.issn.1001-9081.2016.12.3374

Previous Articles     Next Articles

Unvoiced/voiced mode codebook design algorithm based on cellular evenness

XU Jingyun1,2, ZHAO Xiaoqun2, CAI Zhiduan1, WANG Peiliang1   

  1. 1. College of Engineering, Huzhou University, Huzhou Zhejiang 313000, China;
    2. College of Electronic and Information Engineering, Tongji University, Shanghai 201804, China
  • Received:2016-06-01 Revised:2016-07-08 Online:2016-12-10 Published:2016-12-08
  • Supported by:
    This work is partially supported by the National Natural Science Foundation of China (61271248), the Natural Science Foundation of Huzhou (2015YZ04), the Public Welfare Technology Research Industry Project of Zhejiang Province (2016C31115).


徐静云1,2, 赵晓群2, 蔡志端1, 王培良1   

  1. 1. 湖州师范学院 工学院, 浙江 湖州 313000;
    2. 同济大学 电子与信息工程学院, 上海 201804
  • 通讯作者: 徐静云
  • 作者简介:徐静云(1980-),男,江西广丰人,讲师,博士研究生,主要研究方向:语音信号处理;赵晓群(1962-),男,黑龙江依春人,教授,博士,主要研究方向:语音信号处理;蔡志端(1978-),男,江西吉安人,讲师,博士研究生,主要研究方向:故障信号处理;王培良(1963-),男,浙江长兴人,教授,硕士,主要研究方向:故障信号处理。
  • 基金资助:

Abstract: The parameter distribution of unvoiced/voiced Line Spectrum Frequency (LSF) has differences. In order to improve the quantization performance of LSF parameters in vocoder, an unvoiced/voiced mode codebook design algorithm based on Cell Evenness (CE) was presented by using the difference between unvoiced/voiced LSF parameters distribution and CE. Firstly, the optimal amount ratio of unvoiced/voiced LSF parameters participating in the codebook training was deduced according to CE. Then the specified number of atypia LSF parameters were eliminated from unvoiced speech. The final codebook was retrained. The experimental results show that, compared with the shared codebook algorithm under the same bit-rate condition, the average spectrum distortion of the proposed algorithm was reduced by 2.5%, the mean opinion score was increased by 2.3% and the storage of codebook was reduced by 21.1%. The proposed algorithm is also adapted to the vocoder without unvoiced/voiced symbol transmission and the algorithm is also adapted to the vocoder without unvoiced/voiced symbol transmission.

Key words: Line Spectral Frequency (LSF), codebook design, unvoiced/voiced mode, Cell Evenness (CE)

摘要: 清音和浊音线谱频率(LSF)参数分布具有差异性。为了提高声码器中LSF参数的量化性能,利用胞腔均匀度(CE)能定量表征清浊音LSF参数分布的差异程度,提出了一种基于CE的清浊模式码书设计算法。该算法首先根据CE推导出清音和浊音参与训练的LSF参数的数量比;然后剔除清音中指定数量的非典型LSF参数;最后重新训练出码书。实验结果表明,在相同码率情况下,该算法较码书共享算法谱失真降低2.5%,平均意见得分提高了2.3%,码书存储量下降了21.1%,并且适用于不传输清浊音标志的声码器。

关键词: 线谱频率, 码书设计, 清浊模式, 胞腔均匀度

CLC Number: