基于G.729编码参数的语音特征及其应用

目的从低比特率语音编码参数中直接提取语音特征。方法针对G．729编码技术提出了一种从编码参数直接计算倒谱系数和基音／能量轨迹特征的方法。结果该方法通过对残差信号进行线性预测分析，提高了谱包络的精确程度，并从码本增益和延时参数中得到了基音／能量轨迹特征。结论说话人识别的实验结果显示，新方案能够使得基于G．729编码参数的说话人识别效果得到较明显的提高，达到了用解码语音进行识别的水平。

Feature extraction in speaker recognition applications based on G. 729 codec parameters

Aim To extract speech feature directly from codec parameters of low bit rate channels becomes a new trend of speech technology in communication applications.Methods A new scheme of feature extraction is suggested that codec parameters of G.729 is used to analyze residual signal and prosodic information. The LP analysis is used again on residual signal to reconstruct the spectrum fine structure.Results Adaptive codebook delay and two codebook gains are tied to build pitch/energy contour feature.Conclusion When combined with those LPCC derived from LSF parameters, speaker recognition results show that this new feature extraction scheme can distinctly improve the performance, and reach the effect obtained by using synthesized speech.