波形内插语音编码器中快渐变波形幅度的量化

在低速率参数语音编码算法中,如何用有限的比特数有效地量化幅度谱是一个关键问题。本文对波形内插语音编码模型中快渐变波形幅度的量化问题进行了深入研究和分析,提出了一种基于矢量变维和DCT的REW幅度感觉加权量化方案,该方法降低了编码比特数,减少了存储和计算复杂度,增强了编码语音的感性质量。主观听力测试结果表明该量化方案在每帧4比特时的WI语音编解码质量要优于用基于DCT的REW幅度矩阵量化方案在每帧10比特时的重建语音质量。

Quantizaion of REW Magnitude in Waveform Interpolation Speech Coding

How to efficiently quantize magnitude spectra with limited bits is a key issue in low bit rate parameter speech coding. The quantization of rapidly evolving waveform (REW) magnitude spectra in waveform interpolation speech coding is discussed. This paper proposes a quantization scheme with the perceptual weighting for REW magnitude spectra based on variable dimension (VD) of vector and discrete cosine transform (DCT). This kind of method decreases the bit numbers for encoding REW magnitude spectra, reduces storage requirement and computational complexity and enhances the perceptual quality of coded speech. The subjective listening test results show that the WI speech quality using the proposed scheme at four bits per frame for REW quantization exceeds the reconstructed one using the matrix quantized REW magnitude spectra based on DCT with ten bits per frame.