EXPERIMENTAL DESIGN OF SOUND SOURCE LOCALIZATION BASED ON MICROPHONE ARRAYS AND NEURAL NETWORK

This paper presents a sound source localization experimental setup that integrates microphone-array technology with neural networks and the generalized cross-correlation timedelay (GCC-TD) method. The system first preprocesses raw acoustic signals with a Kalman filter, then estimates the time-difference-of-arrival between microphones via GCC-TD, and finally employs a single-hidden-layer neural network to map these time differences onto two-dimensional coordinates, thereby pinpointing the sound source. Replacing conventional mathematical models with a neural network markedly mitigates systematic errors arising from ambient noise, sound-speed variations, and hardware latency, while simultaneously cutting computational complexity and memory requirements. Experimental results demonstrate that the apparatus offers high accuracy, low cost, and real-time display and logging of measurement data.