摘要： 针对低信噪比和噪声变化情况下语音激活检测算法性能下降的问题，提出一种新的参数更新和取值算法。该算法采用Laplacian- Gaussian混合模型对带噪语音频谱的概率分布进行建模，模型参数从带噪语音中估计得到，噪声功率参数通过跟踪语音的音节间隙进行平滑。实验结果表明，该算法在-5 dB信噪比情况下，可以达到95%以上的检测率，具有优越的跟踪性能。

关键词: 语音信号处理, 语音激活检测, 似然比测试, Laplacian-Gaussian混合模型, 噪声跟踪

Abstract: Aiming at the performance decline problem of voice activity detection under low signal-to-noise and noise change circumstances, this paper proposes a new parameter updating and value obtaining algorithm. This algorithm uses Laplacian-Gaussian mixed model, a new, efficient VAD algorithm using Likelihood Ratio Test(LRT) is proposed. The parameters of the model are estimated from noisy voice. The estimated noise power parameter is smoothed during the gaps of voice syllables. Experimental results indicate that the detection rate of this algorithm is above 95% when SNR is -5 dB. This algorithm achieves predominant tracking performance.

Key words: voice signal processing, voice activity detection, Likelihood Ratio Test(LRT), Laplacian-Gaussian mixed model, noise tracking

中图分类号: 

• TN912.3