Sound Context Classification based on Joint Learning Model and Multi-Spectrogram Features

This article presents a deep learning framework applied for Acoustic Scene Classification (ASC), the task of classifying different environments from sounds they produce. To successfully develop framework, we firstly carry out comprehensive analysis spectrogram representation extracted sound scene input, then propose best multi-spectrogram combination front-end feature extraction. In terms back-end classification, novel joint model using parallel architecture Convolutional Neural Network (CNN) and Recurrent (C-RNN), which is able to learn efficiently both spatial features temporal sequences input. The experimental results have proved our proposed general robust ASC tasks by three main contributions. Firstly, most effective indicated specific datasets that none publication previously analyzed. Secondly, CNN C-RNN achieves better performance compared with only baseline in this paper. Finally, competitive state-of-the-art systems on various benchmark IEEE AASP Challenge Detection Scenes Events (DCASE) 2016 Task 1, 2017 2018 1A & 1B, LITIS Rouen.