改进的语音子带清浊音参数量化算法  

Improved Quantization A lgorithm for Unvoiced Voiced Speech Parameter

作　　者：陈亮1 陈亮2 郑静华[3] 张翼鹏[4] 庞亮[5] 

机构地区：[1]解放军理工大学通信工程学院研究生2队,江苏南京210007 [2]解放军理工大学通信工程学院 [3]海军陆战学院模拟训练中心,广东广州510430 [4]南京炮兵学院作战实验中心,江苏南京211132 [5]解放军理工大学通信工程学院研究生1队

出　　处：《军事通信技术》2013年第4期49-53,共5页Journal of Military Communications Technology

基　　金：国家自然科学基金资助项目(61072042)

摘　　要：混合激励线性预测(MELP)算法中,每帧5维的子带清浊音参数对于提高合成语音的自然度有着重要作用,但其每帧5 bit的编码效率给语音的极低速编码带来了困难。文章将MELP的3帧联合构成一个超级帧,对15维的子带清浊音参数进行矢量量化。通过清浊音信息的统计,并利用失真测度d进行码本的优化设计,实现了每个超级帧用3 bit对15维矢量的高效量化。仿真结果表明,文中算法对子带清浊音参数编码后,合成语音仍然保持了良好的可懂度和自然度,可有效应用于600 bps以下极低速语音编码算法中。In MELP algorithm, the sub-band Unvoiced/Voiced（U/V） parameters play an important role in improving the naturalness of synthetic speech. But the coding efficiency with 5 bits per frame brings difficulties for very low bit rate speech coding. In this paper, the three consecutive MELP frames were grouped into a super-frame, and the 15-dim sub-band U/V parameters were quantized. By calculating the U/V probability distribution and optimizing the codebook design by distortion measure d , every 15-dim U/V vector was quantized efficiently with 3 bits for each super-frame. Simulation results show that the intelligibility and naturalness are efficiently maintained for synthesis speech, and the quantization scheme can be applied widely to speech coding algorithm below 600 bps.

关 键 词：混合激励线性预测 子带清浊音 联合矢量量化 语音音质客观评估 

分 类 号：TN912.3[电子电信—通信与信息系统]