基于神经网络的声源定位算法研究

Research on Sound Source Localization Algorithm Based on Neural Network

声音作为人们日常生活中信息表达和接收的重要方式之一,包含了大量的有用信息,其应用范围涵盖了生活的方方面面。因此对于声源定位的深入研究依旧具有广泛的现实意义。本文研究了基于神经网络的声源定位技术,首先介绍了三种传统的声源定位算法,并主要研究了基于相位变换的广义互相关的定位算法(GCC-PHAT),随后根据一种新的单声道和多声道录音数据库(SMARD)建立了房间冲激响应(RIR)模型,并通过麦克风阵列对空间随机生成的声源进行采集,结合全连接神经网络模型进行训练,利用训练好的模型对信号进行分类,最终得到声源的方位。实验结果表明,在不同混响条件和信噪比的环境下,基于神经网络的声源定位算法具有较高的定位准确率。

As one of the important ways of expressing and receiving information in people's daily life,sound contains a lot of useful information,the usage of sound and application scope covers all aspects of life.Therefore,the in-depth study of sound source localization still has broad practical significance.This paper studies the sound source localization based on neural network.First,the paper introduces three traditional sound source localization algorithms,and mainly focus on the Generalized Cross Correlation-Phase Transform(GCC-PHAT)algorithm,and then established a Room Impulse Response(RIR)model according to a new single and multichannel audio recordings database(SMARD).After that with the help of the microphone array,the system collecte sound sources which randomly generated in space,then send the data to a fully connected neural network model for training.Using the well-trained model to classify the signal,and finally get the location of the sound source.The experimental results show that the sound source localization algorithm based on neural network gets a higher localization accuracy under different reverberation conditions and different SNR environments.