一种针对ISF参数的量化算法

宽带语音编码中普遍使用导抗谱频率描述声道。利用转换分类差矢量分裂矢量量化方法对导抗谱频率进行量化，该方法基于转换分类矢量量化及差值分裂矢量量化。IsF矢量先按照给出的码书分类，然后每一类中的差矢量再进行分裂矢量量化。实验结果表明，该算法可在每帧编码比特数为37时达到透明量化要求，并且码书存储量明显少于StephenSo等人给出的转换分类分裂矢量量化方法。

Quantization Algorithm for ISF Parameters

Vocal tract is described by immittance spectral frequencies（ISF） in wideband speech coding. ISF is quantized by using the algorithm of switched difference vector split vector quantization（SDVSVQ）,which is based on switched vector quantiztion and difference vector split vector quantization. ISF vectors are first classified by the given codebook, then difference vectors in every class are splitted and quantized separetely. The experimental results shows that the algorithm could meet the requirement of transparent quantization when the bit rate is 37 bits/frame, and codebook memory requirement is less than switched split vector quantization suggested by Stephen So et al..