应用自相关减法改善噪化语音的线性预测编码分析

本文提出了一种新的线性预测编码(LPC)方法。它既可用于纯语音分析,也可用于噪化语音分析。用于噪化语音分析时,首先在无语音帧内估计出噪声样本的自相关系数,然后从噪化语音的自相关系中减去噪声的自相关系数,最后利用估值得到的语音自相关系数求出它的线性预测系数。当输人噪化语音的信噪比为0～10dB时,使用这种方法可以提高信噪比5dB左右。

Use of Autocorrelation Subtraction Method for Improving LPC Analysis of Noisy Speech

A new linear predictive coding (LPC) method which can be used in quiet as well as noisy environment is studied. When used in the noisy. speech analysis. the noise autocorrelation coefficients are first estimated during the non-speech frames. Then, they are removed by the autocorrelation substraction method from the autocorrelation coefficients of noisy speech. Finally, the linear predictive coefficients of noise are obtained using the estimated speech autocorrelation coefficients. As the signal-to-noise ratio of the input noisy speech ranges from 0 to 10 dB, about 5 dB improvement can be obtained by using this method.