基于线谱对高效矢量量化的0.6kb/s语音编码算法  

Speech coding algorithm at 0.6 kb/s based on efficient vector quantization of linear spectrum pains parameters

作　　者：邹霞[1] 何俊[2] 张雄伟[1] 

机构地区：[1]解放军理工大学通信工程学院,江苏南京210007 [2]通信指挥学院,湖北武汉430010

出　　处：《解放军理工大学学报（自然科学版）》2008年第2期114-118,共5页Journal of PLA University of Science and Technology(Natural Science Edition)

基　　金：江苏省自然科学基金资助项目(BK2006001)

摘　　要：为了提高通信系统的抗干扰和抗攻击能力,尽可能降低语音编码速率。提出了一种0.6 kb/s语音编码算法。算法基于3帧联合,对多帧联合参数采用高效矢量量化,在降低语音编码速率的条件下保证语音编码质量。其中,对线谱对参数采用预测多模式多级矢量量化码本结构。在码本设计过程中,提出了多模式渐进闭环设计,对各类码本联合优化,并联合优化预测器和量化器,可以有效提高线谱对参数量化质量。在译码方,采用多带混和谐波激励提高合成语音清晰度。测试结果表明,该语音编码算法合成语音PESQ(perceptual evaluation of speech quality)得分可以达到2.7,汉语诊断押韵测试DRT(diagnostic rhyme test)得分可以达到89.7。A speech coding algorithm at 0. 6 kb/s was presented to satisfy disadvantaged transmission requirements. The parameters of consecutive 3 frames were grouped together and jointly quantized to improve the speech quality at low bit rate. LSF parameter was quantized with multimode predictive multistage vector quantization and the codebook designed with multimode asymptotically closed-loop design. The novel aspect of the proposed approach was joint optimization of the multimode codebook considering the mode transition between successive superframes. At the decoder, the speech was synthesized with a proposed multiband mixed harmonic excitation model. The results show that the PESQ score can achieve 2.7 points and the DRT score can achieve 89.7.

关 键 词：语音编码 矢量量化 语音合成 线谱对 

分 类 号：TN912[电子电信—通信与信息系统]