一种采用混合激励的超低速率分段声码器  

Mixed-Excited Segment Vocoder at Very Low Bit Rate

作　　者：闵刚[1] 张雄伟[2] 杨吉斌[1] 安云峰[3] 

机构地区：[1]解放军理工大学通信工程学院,南京210007 [2]解放军理工大学指挥自动化学院,南京210007 [3]西安通信学院四系,西安710106

出　　处：《数据采集与处理》2009年第5期680-685,共6页Journal of Data Acquisition and Processing

基　　金：江苏省自然科学基金(BK2006001)资助项目

摘　　要：为满足通信和多媒体存储对超低速语音编码的要求,提出了一种平均编码速率为200和300 bps的超低速率声码器算法。结合分段声码器和M ELP算法的优点,该算法对语音建立了混合激励分段编码模型。提出了线谱对参数的变维矩阵量化和激励参数的变维矢量量化方案,在超低速率条件下获得了较好的量化效果,同时有效地降低了码本存储量。非正式主观听力测试表明:编码速率为300 bps时,重建语音保持了较高的可懂度和一定的自然度;编码速率为200 bps时,语音质量仍可以接受。To meet requirements of communication and multimedia storage, a novel very low bit rate speech coding algorithm at 200 and 300 bps is proposed. Combined advantages of the segment vocoder with MELP algorithm, this algorithm builds a mixed-excited segment encoding model for the speech. Variable dimension matrix quantization （VDMQ） and variable dimension vector quantization （VDVQ） schemes are presented for quantizing LSP and excitation parameters. These schemes have good performance at a very low bit rate. And the codebook storage burden is reduced. Informal subjective listening test shows that the reconstructed speech has high intelligibility and moderate naturalness at 300 bps. The acceptable speech quality is also achieved at 200 bps.

关 键 词：语音编码 混合激励 矢量量化 语音分段 

分 类 号：TN912.3[电子电信—通信与信息系统]