论文部分内容阅读
为在极低速率下实现高质量的语音编码,提出一种高效的带有帧间及级间预测的线谱频率参数多级码本矢量量化(IFP-MSVQ-ISP)算法。算法利用多级矢量量化中上一级码本的选定码矢对残差矢量进行预测,对去除预测分量的残差矢量再进行下一级矢量量化。测试结果表明,这种带有多级码本级间预测的算法与无级间预测的算法相比,能够有效降低线谱频率参数的量化误差,使谱失真降低0.1 dB以上,合成语音客观MOS提高0.02以上。该算法的实现对极低速率下语音压缩编码算法的研究具有重要的参考价值。
In order to achieve high quality speech coding at very low rate, an efficient multi-level codebook vector quantization (IFP-MSVQ-ISP) algorithm with inter-line and inter-level prediction is proposed. The algorithm predicts the residual vector by using the selected code vector of the upper level codebook in the multi-level vector quantization, and then carries out the next-level vector quantization on the residual vector with the predicted component removed. The test results show that compared with the algorithm without inter-stage prediction, this algorithm with multi-level codebook prediction can effectively reduce the quantization error of line spectrum frequency parameters and reduce the spectral distortion by more than 0.1 dB. The synthetic speech objective MOS Increase by 0.02 or more. The implementation of this algorithm has an important reference value for the research of speech compression coding algorithm at very low rate.