The traditional statistical likelihood rate based voice activity detection (VAD) algorithms need to calculate parameters of speech and noise statistical model respectively. Its computational complexity is high. In this paper, we propose to combine cepstrum thresholding estimation method and Rayleigh statistical model for voice activity detection. It uses cepstrum thresholding method to estimate noise spectrum, then take UMPT to get a criterion for updating speech decision threshold based on Rayleigh statistical model. By evaluating 4 different combinational VAD, the simulation results indicate that the proposed method gets a better performance as compared to other combinational VAD method for the non-stationary noise sources.%针对传统的似然比语音活动检测的计算语音与噪声统计模型复杂度高，提出结合倒谱阈值估计噪声频谱与瑞利统计模型的语音活动检测方法。该方法先用倒谱阈值估计噪声的频谱，再利用UMPT获得基于瑞利模型的语音判决阈值更新准则。评估了4种不同方法组合的语音活动检测（voice activity detection，VAD）。实验表明：在非平稳噪声环境下该方法的正确检测率优于其它组合的VAD方法。
展开▼
