倒谱距离和短时能量的语音端点检测方法研究

在讨论传统倒谱距离语音端点检测方法不足的基础上,提出了一种基于倒谱距离和短时能量的语音端点检测改进方法。基于倒谱距离的单参数端点检测方法在高信噪比环境下效果较好,然而在低信噪比的环境下其端点检测性能急剧下降。通过分析倒谱距离和短时能量各自的端点检测特性,建立了一种结合二者特点的双参数判决准则,在保证运算量没有显著增大的前提下提高了端点检测的准确率。仿真实验结果表明,新方法相对于基本倒谱距离端点检测方法,在低信噪比的高斯白噪声环境下端点检测性能有较明显提高。

Study on Speech Endpoint Detection Based on Cepstrum Distance and Short-time Energy

Based on the shortages discussion of traditional cepstrum distance speech endpoint detection method,the speech endpoint detec-tion method based on cepstrum distance and short-time energy is proposed. The endpoint detection method of single parameter of ceps-trum distance is effective in high SNR environment, but the endpoint detection performance falls sharply in low SNR environment. Through the analysis of cepstrum distance and short-time energy endpoint detection features,combined the characteristics of double pa-rameters establish a judgment criteria,which improves the accuracy of endpoint detection under the premise of no significant increase in computational complexity. By comparing the new method with traditional cepstrum distance speech endpoint detection method,simulation experimental results show that the new method for endpoint detection performance is improved obviously under Gaussian white noise in low SNR environment.