The voice activity detection algorithm based on the periodic decomposition of voice was proposed. The traditional voice activity detection algorithms perform poorly in low signal to noise ratio environments of dynamic background noise, because the detective features energy and zero crossing rate, etc, are based on the stationary noise, and are sensitive to the change of signal to noise ratio in the traditional methods. The proposed method is able to solve this problem very well, because the periodicity of the voiced sound part in speech is a very important feature different from the noises and it varies weakly with the type and the signal to noise ratio of the background noise. The accuracy of the pitch extraction has a great influence on the final result of voice activity detection. In view of this situation, the fused pitchextraction method was also proposed, which combined autocorrelation, circular average magnitude difference and YIN. The method obtains better results than any of the three methods with the background noise being white noise, car noise and bubble noise at signal to noise ratios of 0 dB, 5 dB and 10 dB in the accuracy of pitch extraction and voice activity detection.%介绍了基于语音信号周期分解的语音活动检测算法.传统语音活动检测算法在动态低信噪比背景噪声环境下的效果很不理想,这主要是因为传统方法中提取的能量与过零率等检测特征针对的是平稳噪声,对信噪比的变化很敏感.而本文介绍的周期分解语音活动检测方法能较好地解决这个问题,因为语音信号中浊音段的周期性是区别一般噪声信号的重要特征,并且该特征受背景噪声类型和信噪比变化的影响小.在周期分解语音活动检测方法中,基频提取的准确性对最终检测性能有很大影响.针对此情况,提出了自相关、循环均值幅度差分和YIN三种基频提取算法相融合的方法.实验结果表明,在背景噪声为白噪声、汽车噪声、嘈杂人声以及信噪比0 dB.5 dB,10 dB的情况下,该方法相对单一基频提取算法,可以有效提升基频提取与周期分解语音活动检测的准确性.
展开▼
