CLASSIFICATION USING NETWORK-IN-NETWORK BASED CONVOLUTIONAL NEURAL NETWORK

In this paper, we present our entry to the challenge of detection and classification of acoustic scenes and events (DCASE). This paper describes the result of our proposed system for automatic audio scene classification task. Our approach is based on the deep learning method that is adopted from computer vision research field. The convolutional neural network is adopted to solve the problem of audio based scene classification, specifically the architecture of network-in-network is utilized to build the classifier. For the feature extraction part, mel frequency spectral coefficients (MFCC) is used as the input vector for the classifier. Differ from the original architecture of network-in-network, in this work we perform 1-D convolution operation instead of performing 2-D convolution. The classifier is trained using every frames from MFCC feature set, and the results for every frames are then thresholded and voted to choose the final scene label of audio data. The proposed work in this paper shows a better performance of the provided baseline system of DCASE challenge for both development and evaluation dataset.