论文部分内容阅读
为了解决语音参数编码算法中多级矢量量化中码本尺寸过大,存储量过大,导致搜索复杂度大的问题,提出了多级矢量量化中的码本共享的迭代算法。该算法基于多级矢量量化中各级待量化码矢之间的相似性,采用模拟退火算法,通过迭代得出共享变换系数。在1.2 kb/s的正弦激励线性预测声码器中,采用该算法对线谱对参数进行多级矢量量化。测试结果表明:在共享级别选择恰当时,可降低存储量20%,同时重建语音谱失真损失约为0.02 dB,可见该算法可以有效降低码本容量,同时对语音质量影响极小。
In order to solve the problem of large code size and large storage size in multi-level vector quantization in speech parameter coding algorithm, the iterative algorithm of codebook sharing in multi-level vector quantization is proposed. Based on the similarities between the quantized code vectors at various levels in multi-level vector quantization, the algorithm adopts simulated annealing algorithm to obtain the shared transform coefficients through iteration. In the 1.2 kb / s sinusoidal excitation linear predictive vocoder, the algorithm is used to multistage vector quantization of line spectrum parameters. The test results show that at the appropriate sharing level, the storage capacity can be reduced by 20% and the speech spectral distortion loss can be reduced by about 0.02 dB. This shows that the algorithm can effectively reduce the codebook capacity and has very little impact on voice quality.