摘要
针对当前语音激活检测算法在低信噪比和复杂噪声模型的环境下性能损失的问题,提出了一种基于噪声估计的语音激活检测算法,通过对背景噪声进行自适应估计,得到准确的信噪比门限,同时利用估计背景噪声对短时谱进行白化处理,从而使得谱熵判决准则得以适用于复杂噪声模型的环境。实验证明,算法在低信噪比和复杂噪声模型下性能优于G.729B和AMR中的语音激活检测算法。
In the condition of low signal to noise ratio（SNR） and complex noise model,the performance of the voice activity detection（VAD） algorithms always becomes poor.This paper presents a new VAD algorithm based on noise estimation.Through the adaptive noise estimation,the accurate SNR threshold is decided.Then,after applying whitening filter to short time spectra the spectral entropy can be used to VAD algorithm in complex noise model.The experimental results show that the proposed algorithm gets better performance than G.729B and AMR VAD algorithms in the condition of low SNR and complex noise model.
出处
《信息技术》
2011年第10期5-8,共4页
Information Technology
基金
国家自然科学基金项目(60572081)
关键词
语音激活检测
噪声估计
信噪比
谱熵
白化滤波
voice activity detection
noise estimation
signal to noise ratio
spectral entropy
whitening filter
