CNN-Based Acoustic Scene Classification System

• DOI: 10.3390/electronics10040371
• Publication Date: 2021-02-03T16:54:31Z
• Authors: Yerin Lee; Soyoung Lim; Il-Youp Kwak
• Keywords: Feature (linguistics); Convolution (computer science); Harmonic

Acoustic scene classification (ASC) categorizes an audio file based on the environment in which it has been recorded. This long studied detection and of acoustic scenes events (DCASE). presents solution to Task 1 DCASE 2020 challenge submitted by Chung-Ang University team. addressed two challenges that ASC faces real-world applications. One is recorded using different recording devices should be classified general, other model used have low-complexity. We proposed models overcome aforementioned problems. First, a more general was combining harmonic-percussive source separation (HPSS) deltas-deltadeltas features with four models. Second, same feature, depthwise separable convolution applied Convolutional layer develop low-complexity model. Moreover, gradient-weight class activation mapping (Grad-CAM), we investigated what part feature our sees identifies. Our system ranked 9th 7th competition for these subtasks, respectively.