基于深度学习的环境声音识别  

Speaker recognition based on deep learning

作　　者：史秋莹 郑铁然[1] SHI Qiuying;ZHENG Tieran(School of Computer Science and Technology,Harbin Institute of Technology,Harbin 150001,China)

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《智能计算机与应用》2018年第5期34-37,41,共5页Intelligent Computer and Applications

基　　金：国家自然科学基金(91120303)

摘　　要：作为一种感知周围环境十分有效的方法,环境声音识别(Environment Sound Recognition,ESR)被广泛地应用在机器人导航、移动机器人、音频检索、音频取证以及其它基于情景感知、可穿戴的应用中。目前,较为简单的分类器已经大规模应用在ESR问题中,但却不能很好地反映和识别环境声音,而深度神经网络作为一种高性能、多层的神经网络,为更好地描述原始数据特征和解决模式识别问题提供了更有效的途径。为此,本文将深度神经网络应用在环境声音识别问题中,并对音频特征进行特征融合,通过训练深度信念网络(Deep Belief Network,DBN)进行环境场景的识别。实验结果表明,基于特征融合的方法的识别性能相较于分别采用音频特征的方法识别在性能上有明显提升,且将深度学习应用在ESR问题中具有明显的优势。Environment Sound Recognition（ ESR） is an efficient way to perceive surrounding scenes,which is widely used in many application scenarios like robotic navigation,mobile robots,audio retrieval,audio forensics and other wearable,context-aware applications. Classifiers used in most ESR problems is too simple to express features and classify the environment sounds accurately.Deep Neural Network（ DNN） is a multilayer,efficient neural network,which also can provide a better way for describing features and solving pattern recognition problems. In this paper,deep learning is used in ESR problem,and for sufficient using audio features,different audio features are fusioned by feature-fusion method. Based on the above,the paper classifies the environment sound by training the Deep Belief Network（ DBN）. Experimental results showthat DBN and feature-fusion method could achieve better performance.

关 键 词：环境声音识别 特征融合 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]