摘要
为了解决传统方法在强噪声环境下,语音检测性能急剧下降的缺陷,提高信号在低信噪比(0 db以下)语音端点检测的准确性,本文提出了一种将多窗谱估计谱减法和自适应子带能熵比相结合的检测算法。该算法利用增益因子可变的多窗谱估计谱减法对低信噪比信号进行降噪,提高其信号的信噪比,再将每帧信号分为若干个子带(其数量可自适应选择),提取每个子带能熵比参数进行端点检测。实验结果表明,当信噪比为-10 db时,信号检测准确性维持在95%左右。该方法能在低信噪比情况下,显著提高端点检测准确性和可靠性。
Under the strong noise condition, the voice detection performance decrease dramatically by traditional way. In order to improve the voiee deteetion aeeuracy for the poor signal - to - noise ratio （lower than 0 db） under the strong noise circumstance, this article introduces a new method which is based on spectral subtraction of mlutitaper spectral estimation and adaptive sub - band energy - entropy - ratio. This algorithm is using gain taetor of variable multitaper spectral estimation to denoise the poor signal and improve the signal - to - noise ratio, then divide the signal frame into several sub - bands whose quantity is adaptive choosing. Abstract the data from the sub - band energy - entropy - ratio to detect the endpoint. The experiment result shows that the signal detection accuracy is around 95% when the SNR of voice signal is lower than 0 db. The method of this paper improves accuracy and rcliability of lhc endpoint detection under lhc lower SNR circumstance.
作者
陈莹莹
毕春艳
龙建忠
CHEN Yingying;BI Chunyan;LONG Jianzhong(School of Electrical and Electronic Information Engineering,Jinjiang College,Sichuan University,Pengshan 620860,China;School of Electronics and Information Engineering,Sichuan University,Chengdu 610065,China)
出处
《电视技术》
2018年第6期9-12,27,共5页
Video Engineering
基金
四川省教育厅科研项目(17ZB0261)
关键词
低信噪比
多窗谱
谱减法
子带能熵比
自适应算法
端点检测
Low signal - to - noise ratio
muhitaper method
spectral subtraction
sub - band energy- entropy - ratio
adaptive algorithm
endpoint detection
