Sound Event Detection Based on Beamformed Convolutional Neural Network Using Multi-Microphones

摘要

This paper presents a real environment sound event detection method based on pre-processing technology. Our goal is to improve the performance of the sound event detection using a pre-processing module called parameterized multi-channel non-causal Wiener filter (PMWF). First, we convert the existing 1 channel data to 2 channels through the Room impulse response generator (RIR) module. The reason for 2-channel conversion is that PMWF requires multiple channels for beamforming. Noise cancellation is performed through PMWF and the results are derived through the proposed convolutional neural network model. As a result, we found that this method has a good effect on real-time sound event detection, and we found that peak normalization and median filter also have a good effect.