Acoustic Scene Classification Using A Deeper Training Method for Convolution Neural Network
Tan Nhu Nhat Doan, Hung Nguyen, Dat Thanh Ngo, Lam Dang Pham, Ha Hoang Kha · 2019
In this paper, we present a deep learning framework applied for acoustic scene classification (ASC) recognizing the environmental sounds. Since an audio scene related to a given location potentially contains numerous sound events, only few of these events supply helpful information on the scene, which makes the acoustic scene classification task become a very complex problem. To confront this challenge, we suggest a novel architecture consisting of two basic processes. The front-end process approaches a spectrogram feature, using Gammatone filters. Regarding the back-end classification, we propose a novel convolutional neural network (CNN) architecture that enforces the network deeply learning middle convolutional layers. Our experiments conducted over DCASE2016 task 1A dataset offer the highest classification accuracy of 84.4% as compared to 72.5% of DCASE2016 baseline.