一种低信噪比环境下的语音端点检测算法  

A Speech Endpoint Detection Algorithm in Low Signal to Noise Ratio Environments

作　　者：陈锡锻 张恪莱 CHEN Xiduan;ZHANG Kelai(Zhejiang Industry&Trade Vocational College,Wenzhou 325000,China)

机构地区：[1]浙江工贸职业技术学院,浙江温州325000

出　　处：《电声技术》2023年第4期66-68,共3页Audio Engineering

基　　金：2022年杭州钢铁集团有限公司项目:基于HMM模型的钢铁生产设备故障检测方法研究。

摘　　要：为了改善在低信噪比背景下子带谱熵语音端点检测算法的性能,提出了一种多正弦窗子带谱熵检测算法。首先介绍了目前比较流行的子带谱熵语音端点检测算法的原理,其次分析了多正弦窗口谱估计的特点,最后将多正弦窗口谱估计引入到子带谱熵检测中形成多正弦窗子带谱熵这一新的特征,利用多正弦窗口谱估计具有更小的方差和更高的分辨率来改善低信噪比环境下子带谱熵检测算法的性能。通过在不同信噪比环境下仿真验证,证明了相对于子带谱熵检测算法,所提算法在低信噪比环境下拥有更高的检测准确率和更好的噪声鲁棒性。In order to improve the performance of sub-band spectral entropy speech endpoint detection algorithm in the background of low SNR,a multi-sine window band spectral entropy detection algorithm is proposed.Firstly,the principle of the popular sub-band spectral entropy speech endpoint detection algorithm is introduced,then the characteristics of multi-sine window spectral estimation are analyzed.Finally,multi-sine window spectral estimation is introduced into sub-band spectral entropy detection to form a new feature of multi-sine window band spectral entropy,and the performance of sub-band spectral entropy detection algorithm is improved by using multi-sine window spectral estimation with smaller variance and higher resolution.The simulation results in different SNR environments show that this algorithm has higher detection accuracy and better noise robustness than the subband spectral entropy detection algorithm in low SNR environment.

关 键 词：低信噪比 子带谱熵 多正弦窗谱估计 准确率 鲁棒性 

分 类 号：TN912.3[电子电信—通信与信息系统]