低速率声码器激励谱参数的估值算法

The Statistic-Based Prediction of Excitation Spectral Parameters in Low Bit-rate Vocoder

在声码器算法中,对激励信号的描述直接影响了重建语音的质量。在600bps及以下速率的一些声码器中,因为没有足够的编码比特,往往对激励谱幅度参数不进行编码传输,而直接用全1表示。为了改善重建语音的质量,本文根据激励谱幅度参数与其他参数间的相关性,提出了2种码本映射的算法,在接收端利用其它参数对激励谱幅度参数进行估值,以减少采用全1激励谱幅度参数的误差。将其应用在正弦激励线性预测(SELP)600bps声码器中进行测试,2种算法分别能减小激励谱幅度参数的失真23%和26%,使合成语音的MOS分提高1%和1.1%。

In the algorithm of vocoders, the description of excitation signal has a great impact on the quality of synthesized voice. In 600bps and lower bit-rate vocoders, in order to save the encoding bits used for parameters, the excitation spectral parameters are not transmitted, and instead, replaced by 1. In order to improve the quality of synthesized voice, an algorithm of codebook mapping based on the statistics and the correlation between excitation spectral parameters and other parameters, is presented in th...