【摘要】 针对传统的信源优化多级矢量量化抗误码性能较差的问题,提出了一种结合信源和信道的多级矢量量化码本联合优化算法。该算法将码本联合优化与非等重信道保护相结合,充分利用多级矢量量化中各级码字之间的相互作用关系和非等重信道保护的特性,对各级码字进行非等重误码率的迭代优化来降低整个系统失真。在低速率语音编码中线谱频率参数的仿真测试表明:与信道优化的多级矢量量化独立码本和非等重信道保护相结合的方案相比,在8%误码率信道下该算法线谱频率参数的平均谱失真降低了0.1dB;与等重误码率的码本联合优化方案相比,在各种误码率信道下该算法线谱频率参数的平均谱失真都有明显降低。更多还原

【Abstract】 Source optimized multi-stage vector quantization can have poor quality due to channel errors.This paper presents a joint optimization algorithm combining source and channel coding for multistage vector quantization.The joint optimization of the codebook with unequal error protection optimizes each codeword in each iteration using unequal bit error rates to reduce the distortion in the whole system by taking advantage of the connections between each codeword in each stage and the characteristics of unequal error protection.Simulations using linear spectral pairs of low bit speech coding show that the average spectral distortion is reduced 0.1dB for an8% bit error rate compared to combined channel optimized multistage individual vector quantization and unequal error protection.The average spectral distortion is much less than with using multistage codebook joint optimization of equal bit error rates.更多还原