Parallel Capsule Neural Networks for Sound Event Detection

Abstract

In this work, we propose a sound event detection system based on a parallel capsule neural network. The system takes advantage of the capability of capsule neural networks in the detection of overlapping objects. It further develops a parallel architecture and uses the kernel design of different shapes and sizes to effectively utilize the feature information to increase the detection accuracy. The experimental results show that the performance of the proposed system is as low as 52.34% measured by the error rate, which is even lower than the rank 1 system in DCASE2017 challenge.