基于AMR编码参数的语音识别

Speech Recognition Based on AMR Vocoder Parameters

基于语音编码系统的语音识别 ,由于受编码的影响其识别效果在编码速率下降时显著降低。传统的识别方法从重构语音波形中提取特征参数 ,并针对该特征参数进行训练和识别。比较了基于编码语音的识别准确率和基于编码参数的识别准确率 ,并研究了编码参数对识别准确率的影响。在此基础上 ,通过选择受编码影响较小的编码参数 ,直接将 LPC参数和残差信号参数组合起来构成特征参数进行语音识别。实验结果表明 ,采用这种方法的 AMR语音识别系统 ,其识别效果接近于基于原始语音的识别效果。

Speech coding affects speech recognition performance by deteriorating recognition accuracy as the coded bit rate decreases. The conventional systems that recognize coded speech reconstruct the speech waveform from the coded parameters and then perform recognition based on the characteristic parameters of the waveform. In this paper, a comparison is made between the recognition accuracy of coded speech and the accuracy obtained when using the features derived from the coding parameters. The effects of coding on the recognition accuracy is analyzed. The cepstral streams representing the LPC parameters are combined with residual parameters to recognize directly from the coded parameters. Experiment results suggest that it is possible to obtain recognition accuracy equal to the conventional systems from reconstructed waveforms.