基于矢量量化的语音信号频带扩展

Bandwidth Expansion of Speech Based on Vector Quantization

对基于矢量量化的频带扩展方法进行了改进.在码本形成上提出了重新量化的方法,并用码本结合浊音度的方法调整增益.首先根据清浊度和能量被划分为标准将窄带输入信号清音、浊音和静音3类;然后每类信号选择不同的码本,用基于矢量量化的方法将窄带信号的谱包络转换成高频带信号的谱包络;再用激励信号(高斯白噪声信号)和重建的高频谱包络合成高频带语音;最后将高频带与原窄带信号之和作为最终的宽带信号.仿真及与其他方法比较说明,本文的方法所需计算量小,适合实时环境.

A progress on the traditional technique for high-band spectral envelope prediction based on codebook mapping is presented. A re-quantize method to train the codebook and a new refinement method based on VQ and voicing degree are also proposed. First, the narrow band speech is divided into three groups by voicing degree and energy, and in each group, the high-band spectral envelope is determined from the high-band code vector which is closest in shape to the spectral envelope of the frame of input narrowband speech under analysis. Second, the Gaussian white noise is adopted as the excitation signal to synthesis the high band signal. The method mentioned here is proved suitable for real time signal processing.