DCASE 2018 Challenge baseline with convolutional neural networks
share
The Detection and Classification of Acoustic Scenes and Events (DCASE) is a well-known IEEE AASP challenge consisting of a number of audio classification and sound event detection tasks. DCASE 2018 has five tasks: 1) Acoustic scene classification, 2) General-purpose audio tagging, 3) Bird audio detection, 4) Weakly-labeled semi-supervised sound event detection and 5) Multi-channel audio tagging. In this paper, we describe our Python implementation of all of these tasks, which we have released under an open-source license. The baseline source code contains the implementation of convolutional neural networks (CNNs), including AlexNetish and VGGish -- networks originating from computer vision. We researched how the performance varies from task to task with the same configuration of neural networks. Experiments show that the deeper VGGish network performs better than AlexNetish on all tasks except Task 1, where VGGish and AlexNetish perform similarly. With the VGGish network, we achieve an accuracy of 0.680 on Task 1, a mean average precision (mAP) of 0.928 on Task 2, an area under the curve (AUC) of 0.854 on Task 3, a sound event detection F1 score of 20.8% on Task 4 and an F1 score of 87.75% on Task 5.
READ FULL TEXT
