摘要
为了解决低信噪比条件下现有语音活动检测算法可靠性难以保证的问题,提出长时子带能量变化特征,度量短时谱子带能量沿时间轴的变化程度。基于TIMIT语音库采用高斯混合模型评价所提特征的性能。实验结果表明,在五种噪声的不同信噪比条件下,提出的语音活动检测算法性能优于传统的VAD。
Concerning the issue that the reliability of the current voice activity detection （VAD） algorithm is difficult to guarantee at low signal-to-noise ratio （SNR） conditions, this paper presented the measure of Iong-term sub-band energy variability to capture the sub-band energy of short-time spectrum varying over time. The performance of the feature was evaluated using Gaussian mixture models （GMMs） on the TIMIT corpus. Experimental results showcd the accuracy of the proposed VAD scheme was better than that obtained by the traditional VAD schemes under five types of noises and different SNR conditions.
出处
《移动通信》
2016年第14期25-28,共4页
Mobile Communications
关键词
语音活动检测
长时子带能量
高斯混合模型
voice activity detection long-tem sub-band energy Gaussian mixture models （GMMs）
