摘要
本文在分析基本谱熵端点检测算法局限的基础上,引入基于二阶累积量的门限更新方法,加入短时能量参数,提出基于加权谱熵的检测方法;此外,引入特征空间能量熵定义,建立新的门限确定准则,提出基于特征空间能量熵的检测方法。通过对平稳高斯白噪声、M109坦克噪声和F16战斗机噪声这三种典型噪声环境下信噪比(SNR)从-5dB到20dB的带噪语音信号进行的仿真实验分析表明,所提两种方法能更为准确地检测到语音的端点。
On discussing the defects of the basic spectral entropy based voice activity detection method, we propose two improved projects. One method based on spectral entropy and short-time energy is established by introducing threshold updating method based on second-order cumulants ;The other one based on eigenspace-energy-entropy is established by improved threshold ascertaining rules. We make simulation with a lot of noisy speech signals ranging from -5dB to 20dB which is in three representative noisy environments. The simulation results show that the comparatively accurate speech endpoints can be detected by the improved projects proposed before in different SNR environments.
出处
《信号处理》
CSCD
北大核心
2009年第3期368-373,共6页
Journal of Signal Processing
基金
陕西省自然科学基金(2006F40)资助课题
关键词
端点检测
熵
判决准则
语音增强
Voice activity detection
Entropy
Discriminative rules
Speech enhancement
