基于机器学习的单声源定位算法

Single Sound Source Localization AlgorithmsBased on Machine Learning

研究基于麦克风阵列模型单声源定位机器学习算法,首先构建基于单点声源—三点接收模型,对声波信号进行采集、处理,对线性回归、决策树、多层感知机以及集成学习的原理进行详细分析,并将其应用于声源定位模型中,使用Python机器学习工具箱sklearn对算法进行实现,进一步比较及评估不同算法模型的预测结果和性能指标。针对实测数据的仿真结果,对4种算法各自特点进行了分析,为不同应用环境和场合下声源定位算法的选择提供了一定的参考。

The single-source localization machine learning algorithm based on a microphone array mod⁃el was studied.Firstly,a model based on single-point sound source and three-point receiver was built,the related signals were collected and processed.Then,the principles of linear regression,decision tree,multilayer perceptron,and integrated learning were analyzed in detail,the above-mentioned four sound source localization algorithms were implemented by using sklearn.Furtherly,the prediction results and performance indicators of the different algorithm models were compared and evaluated.Based on the ac⁃tual simulation results,the characteristics of the four algorithms were analyzed and summarized,which provides a reference for the selection and study of sound source localization algorithms in different ap⁃plication environments and occasions.