Feature Extraction and Enhanced Classification of Urban Sounds

Asma Begum, Afshaan Kaleem

Abstract: Urban Sound Classification is an important but challenging problem. In this paper, we present a new deep convolutional neural network for classification tasks that combines MFCC with Mel spectrogram. In comparison to using a single feature, this feature combination can make the features richer. The network suggested extracts and derives high-level features using three convolutional blocks, each of which is made up of two convolutional layers and a pooling layer. We apply a filter with a limited receptive field in each convolutional layer to preserve the network's depth and lower the number of parameters. On ESC-50 and UrbanSound8K, where our technique was tested, classification accuracy was 45.60% and 91.0%, respectively. The experimental results show that the proposed method is suitable for Urban Sound classification

Keywords: MFCC, Feature Extraction. Deep learning, Urban Sound classification