一种改进的语音信号端点检测方法研究

在语音识别系统中端点检测有误差会降低系统的识别率,进行有效准确的端点检测是语音识别的重要步骤.当信噪比较低时,传统的端点检测方法不能有效的工作.为了提高系统的识别率,本文提出了一种更有效的端点检测算法,基于LPC美尔倒谱特征的端点检测方法.它是基于倒谱特征方法的一种改进.实验证明,该算法在低信噪比的情况下,能够准确的检测出语音信号的端点.通过对三种不同的端点检测算法的比较,证明了基于LPC美尔倒谱特征算法在低信噪比的情况下有较高的检测正确率.

Improved method for speech endpoint detection

The correct rate of speech recognition will go down severely for errors from the inaccurate endpoint detection. Endpoint detection is an important step in speech recognition. Some traditional detection algorithms cannot take effect and reduces the recognition as low SNR. While an algorithm for speech endpoint detection based on LPC Mel cepstral feature designed to cover this disadvantage is proposed in this paper. The experiments show that this algorithm can detect speech endpoint accurate as low SNR. It proves that the method has better detection capability with low SNR compared with other two endpoint detection algorithms.