基于二维PMCC鲁棒特征参数的语音识别

在噪声环境语音识别中，如何提取鲁棒性特征参数是其核心问题之一，首先提出了一种二维根倒谱特征参数，然后，该参数结合基于最小方差无失真响应谱估计的特征参数(PMCC)。最终，发现了一种新颖的鲁棒特征参数，在不同的信噪比下，它能成功地被用于连续语音识别中。试验结果表明，在不同的噪声环境和信噪比下，二维PMCC鲁棒特征参数比传统Mel频率倒谱系数（MFCC）和感知线性预测（PLP）有更好的识别率。

Speech recognition based on two-dimensional PMCC robust feature parameter

One of the key problems in noise speech recognition is how to extract the robust feature parameters. A two-dimensional root cepstrum feature parameter was first proposed, and then this parameter was combined with robust feature based on the Minimum Variance Distortless Response (MVDR) method of spectrum estimation proposed. Finally, a novel robust feature was presented to be successfully used in continuous speech recognition under different SNRs. Experimental results indicate that the two-dimensional PMCC robust feature parameter is superior to conventional Melt Frequency Cepstral Coefficients （MFCC） and Perceptually Linear Prediction （PLP） in improving the recognition accuracy under different noise conditions and SNRs.