基于小波变换的2.4kbit/s波形内插语音编码算法

Waveform interpolation speech coding algorithm at 2.4kbit/s based on wavelet transform

基于双正交小波滤波器组对波形内插编码中提取的特征波进行多级分解与重构,提出了一种基于小波变换(WT)的2.4kbit/s特征波形内插(CWI)语音编码算法。编码端去除了特征波对齐运算,并对幅度谱进行多级分解,相位谱不传输,鉴于小波变换对信号的压缩特性,仅传输对人耳感知起主要贡献的最后一级特征波幅度谱;解码端对各尺度空间采用单独重建的方法,相位信息在重构的末级与幅度谱结合,并由浊音度标志选择固定或随机相位。此外,根据语音信号的时变特性,由基于子帧的浊音度标志选择需要传输的幅度谱及量化模式。主观R-A/B测试表明,这种基于小波变换的2.4kbit/s编码算法的合成语音质量明显优于标准的2.4kbit/s的MELP编码器及FS1016的4.8kbit/sCELP编码器,亦优于3.8kbit/s的传统CWI编码框架下的合成语音效果。

A waveform interpolation speech coding algorithm at 2.4kbit/s based on wavelet transform was proposed, in which characteristic waveform decomposition and reconstruction is processed using biorthogonal wavelet filters. At the encoder, the alignment operation was removed and the amplitude spectrum is decomposed without phase spectrum. Because of the wavelet compression, only the amplitude spectrum of last layer is transmitted. At the decoder, different scale space is reconstructed separately and random or fixed phase was combined to the spectrum based on voice degree at the last. Additionally, based on the time-variable feature of speech, voiced degree flag of each subframe was introduced to indicate the spectrum selection and quantization. Subjective R-A/B listening tests indicated that the synthesis speech quality of the 2.4kbit/s WT-CWI coding algorithm is better than that of standard 2.4kbit/s MELP coder and FS1016 4.8kbit/s CELP coder and is better than traditional 3.8kbit/s CWI coder.