论文部分内容阅读
矢量量化可有效降低语音编码速率,但目前已有的多级分裂矢量量化、转换分类分裂矢量量化方法等都存在存储需求、计算复杂度以及解码语音质量等不能达到良好折衷的缺陷。该文提出了一种码书分类重排矢量量化方法。该方法通过将设计好的码书进行分类重排以降低码书搜索范围。并将该方法与多级分裂矢量量化结合,提出了码书分类重排多级分裂矢量量化方法。在量化比特及码书大小不变的前提下,实验结果表明:该方法可达到透明量化效果,量化时的计算复杂度最大降幅可达到多级分裂矢量量化方法的90.24%。
Vector quantization can effectively reduce the speech coding rate, but the existing multi-level splitting vector quantization, splitting and splitting vector quantization methods have the storage requirements, computational complexity and decoding speech quality can not achieve a good compromise flaw. In this paper, we propose a new method of vector quantization of codebook classification and rearrangement. The method reduces the codebook search range by classifying and rearranging the designed codebooks. Combining this method with multistage splitting vector quantization, we propose a new method for vector quantization of codebook classification and rearrangement. Experimental results show that this method can achieve the effect of transparent quantification with the quantization bit and codebook size unchanged, and the maximum decrease of the computational complexity in quantization can reach 90.24% of the multi-level splitting vector quantization method.