论文部分内容阅读
为降低在窄带高误码率信道上语音传输的失真,针对线谱对参数提出了一种基于信源信道联合特性的声码器线谱对参数抗误码算法。该算法首先对声码器模式信息采用结合长时统计特性的改进最大后验概率算法进行恢复,然后对语音合成质量影响最大的线谱对参数进行基于前向统计概率和分模式加权的最小均方误差准则下的线谱对参数差错后处理。该算法在不消耗任何额外带宽且无算法延时的要求下,可以显著降低线谱对参数谱失真,提高声码器抗信道误码能力和信道误码条件下的合成语音质量。特别在较高信道误码率(5%)下,平均谱失真降低0.3,MOS(meanopinionscore)平均提高15%左右。
In order to reduce the distortion of speech transmission on narrowband high bit error rate (SNR) channel, a parametric error resilience algorithm is proposed for vocoder parameters based on joint feature of source channel. Firstly, the vocoder mode information is restored by using the improved maximum a posteriori probability algorithm combined with long-term statistics, and then the line spectrum with the strongest impact on the speech synthesis quality is minimized based on the forward statistical probability and the sub-mode weighting The line spectrum under the square error criterion is the parameter error post-processing. The proposed algorithm can significantly reduce spectral distortion of line spectrum and improve the quality of vocoder’s anti-channel error ability and synthetic speech quality under the condition of channel error without consuming any additional bandwidth and without algorithm delay. Especially at the higher channel error rate (5%), the average spectral distortion decreases by 0.3, meanwhile the mean increase in MOS (MOS) is about 15%.