基于谱减与自适应子带对数能熵积的端点检测  

Endpoint detection based on spectral subtraction and logarithmic energy entropy product of adaptive sub-bands

作　　者：张洪德 韩鑫怡 柳林 柳扬 ZHANG Hongde;HAN Xinyi;LIU Lin;LIU Yang(Communication Sergeants College, PLA Army Engineering University, Chongqing 400035, China;Hefei Flytek Digital Technology limited company, Hefei 230088, China)

机构地区：[1]陆军工程大学通信士官学校,重庆400035 [2]合肥讯飞数码科技有限公司,合肥230088

出　　处：《兵器装备工程学报》2022年第2期267-273,共7页Journal of Ordnance Equipment Engineering

基　　金：军内科研项目(LJ20191C070659)。

摘　　要：针对低信噪比环境下语音端点检测准确率低、鲁棒性差,提出了一种将谱减降噪和自适应子带对数能熵积相结合的语音端点检测算法。首先利用改进的多窗谱估计谱减法提升语音信号质量,再以自适应子带对数能熵积这一新的语音特征参数为阈值,使用动态阈值双门限检测法进行语音端点检测。实验结果表明,该算法针对低信噪比语音信号具有更好的准确性和鲁棒性。Aiming at the problems of low accuracy and poor robustness of speech endpoint detection in low SNR environment,a speech endpoint detection algorithm combining spectral noise reduction and the product of logarithmic energy entropy of adaptive sub-bands was proposed.The improved multi-window spectral subtraction method was used to improve the quality of speech signals.A new speech feature parameter,the logarithmic energy entropy product of adaptive sub-band,was used as the threshold,and the dynamic threshold double-threshold detection method was used to detect speech endpoints.Experimental results show that the proposed algorithm has better accuracy and robustness for low SNR speech signals.

关 键 词：语音端点检测 子带对数能量 子带谱熵 多窗谱估计谱减法 双门限检测 

分 类 号：TN912.3[电子电信—通信与信息系统]