基于条件PDF的宽带ISF参数分裂矢量量化方法

Split vector quantizer for wideband ISF parameters based on conditional PDF

宽带ISF参数的矢量量化是语音编码中的重要环节,其量化性能的高低对于解码端语音的质量有重要影响.针对宽带语音ISF参数矢量量化问题,提出了一种新的量化方法.该方法利用ISF参数帧间相关性,将相邻2帧ISF参数的条件PDF用高斯概率模型表示.与传统分裂矢量量化不同,该方法首先根据前一帧的量化结果对当前帧分类、选择合适的码书,然后对该帧在选定的码书中进行分裂矢量量化.实验表明,该算法在每帧编码比特数44时达到透明量化,且平均谱失真比利用传统分裂矢量量化时的谱失真小.

Vector quantizing of wideband ISF parameters is an important part of speech coding,and its performance decides the quality of the speech after decoded.A new vector quantization method is proposed for coding the wideband ISF（Immittance Spectral Frequencie）sparameters.In this approach,the conditional PDF（probability density Function）for the parameter vectors of successive source frames is modeled using a GMM（Gaussian mixture mode）l because there exists interframe correlation.In this quantizer,the current frame is classified to choose appropriate quantization codebook based on the previous frame firstly,and then the ISF parameter is quantized using split vector quantization.Experimental results show that this quantization scheme achieves transparent coding at 44bits/frame,and the average spectral distortion is lower than that of traditional split vector quantization.