多语音特征参数的端点检测方法研究

摘要

为了提高语音端点检测算法的鲁棒性,提出了一种在不同信噪比下采用不同语音特征参数的端点检测算法.对含噪语音进行基于背景噪声能量估计的信噪比估计,根据估计的信噪比大小选择不同的特征参数来进行端点检测,在高信噪比下采用传统的语音短时能量和过零率,在低信噪比下采用基音周期、高频与全频带能量比和谱失真,即算法能根据信噪比的大小来自适应调整检测方法.实验结果表明,该方法具有良好的鲁棒性,在不同的信噪比下检测的准确率都很高.%In order to improve robustness of voice activity detection (VAD) algorithm, a VAD algorithm is presented with different voice characteristic parameter under different signal-to-noise ratio (SNR). First SNR of noisy voice is estimated based on background noise power estimation, then VAD is enabled with different characteristic parameter under different SNR Voice energy and zero cross ratio is used to high SNR, while under low SNR, pitch period, high band to full band energy ratio and spectral distortion is used, so the algorithm can adjust detection method according to SNR adaptively. Simulation results show that the algorithm have better robust, and its accuracy is very high under different SNR