Indoor Sound Source Localization Algorithm Based on BP Neural Network

Sound source localization is an important part of the perception of things around. Sound source localization can overcome the shortcomings of visual localization, and can also locate the invisible place. The application of sound source localization in indoor is the latest trend. Starting from the application of deep learning to indoor sound source localization, this paper focuses on the analysis and research of BP neural network applied to indoor sound source localization algorithm. In this paper, an off-line sampling scheme is used to construct the network structure with 7 neurons hidden in the layer, and the BP algorithm of LevenBerg-Marquardt is used as the training function, this algorithm can solve the traditional algorithm through the study of the physical properties of sound, set up the corresponding equation, and then solve, the process is complex, to solve the difficult problem. The simulation results show that the algorithm can be implemented in 100 square meters of the house, through sampling 400 sets of data for machine training, positioning error can be controlled in a few centimeters effect.

声源定位是对周围事物感知的重要部分。声源定位能够克服视觉定位的缺陷，还能对看不见的地方进行定位。声源定位在室内的应用是最新的趋势。本文从深度学习在室内声源定位中的应用出发，重点对应用于室内声源定位算法的BP神经网络进行分析和研究。本文采用离线采样方案构建了隐藏层有7个神经元的网络结构，并采用LevenBerg - Marquardt的BP算法作为训练函数，该算法通过对声音物理特性的研究来解决传统算法需建立相应方程然后求解，过程复杂、求解困难的问题。仿真结果表明，该算法在100平方米的房屋内可实现，通过对400组数据进行采样用于机器训练，定位误差可控制在几厘米的效果。