To achieve good reconstruction speech quality in very low bit rate speech codecs, an efficient dimension reduction quantization acheme for linear spectrum pair (LSP) parameters was proposed based on compressed sensing. In the encoder, the LSP parameters extracted from consecutive speech frames are shaped into a high dimensional vector, and then the dimension of the vector is reduced by CS to produce a measurement vector, the measurements are quantized using the split vector quantizer. In the decoder, according to the quantized measurements, the original LSP vector is reconstructed by the orthogonal matching pursuit method. Experimental results show that the scheme is more efficient than that of conventional matrix quantization scheme and DCT based dimension reduction quantization scheme, the average spectral distortion reduction of up to 0.23dB and 0.13dB is achieved respectively. Informal subjective listening test shows that the reconstructed speech has moderate intelligibility and naturalness, it is observed that the degradation in speech quality is tolerable and with low codebook storage requirements.%为实现高质量的极低速语音编码,提出一种基于压缩感知理论的线谱对(LSP)参数降维量化算法.编码端利用压缩感知理论对超帧LSP高维矢量进行降维处理,将原始LSP参数投影到低维空间,得到低维测量值,然后采用分裂矢量量化算法对测量值进行量化;解码端以量化后的测量值为已知条件,利用正交匹配追踪算法重构出原始LSP高维矢量.实验结果表明,本算法相对低速语音编码中的矩阵量化方案,平均谱失真降低了0.23dB,相对基于DCT变换的降维量化方案,平均谱失真降低了0.13dB.这种先降维再量化的思想可以大幅减少编码所需的比特数及码本存储复杂度,有效降低语音编码速率,并且合成语音可懂度、自然度较高,音质虽有所失真,但基本上感觉不到明显的听觉质量下降.
展开▼
