利用能量压缩后的MBPD检测低信噪比声音事件  

DETECTION OF LOW SNR SOUND EVENTS USING MBPD AFTER ENERGY COMPRESSION

作　　者：林艺明 李应[1] Lin Yiming;Li Ying(College of Mathematics and Computer Science,Fuzhou University,Fuzhou 350108,Fujian,China)

机构地区：[1]福州大学数学与计算机科学学院,福建福州350108

出　　处：《计算机应用与软件》2021年第6期126-133,共8页Computer Applications and Software

基　　金：国家自然科学基金项目(61075022);福建省自然科学基金项目(2018J01793)。

摘　　要：针对低信噪比下的声音事件检测问题,提出基于能量压缩和灰度增强的多频带能量分布图的声音事件检测方法。将声音数据的gammatone频谱转成能量谱,对不同频带的能量进行不同比例的能量压缩,计算其多频带能量分布图,并对其进行灰度增强;对调整后的多频带能量分布图进行8×8的分块,对每一子块进行奇异值分解,提取主要数值作为声音事件的特征;利用随机森林分类器对特征建模与检测。实验结果表明,在低信噪比环境下,该方法具有良好的检测效果。Aiming at the problem of sound event detection in low SNR,a multi-band energy distribution map based on energy compression and gray enhancement is proposed.The gammatone spectrum of sound data was converted into power spectrum,and the energy of different frequency bands was compressed in different proportions;the multi-band power distribution map was calculated and the gray level was enhanced;the adjusted multi-band power distribution map was divided into 8×8 blocks,each sub-block was decomposed by singular value decomposition,and the main values were extracted as the features of sound events;the random forest classifier was used to model and detect features.The experimental results show that the method has good detection effect in low SNR environment.

关 键 词：声音事件检测 多频带能量分布 能量压缩 灰度增强 奇异值 随机森林 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]