Article Submission

Publisher

View Articles


Guidelines for Authors

Abstracting/Indexing

Order Journal
Volume 1, Issue 1, January 2021

Original Research


Music Genre Classification using Neural Networks with Data Augmentation

Macharla Vaibhavi and P. Radha Krishna

Department of Computer Science and Engineering National Institute of Technology Warangal, India

J.Innovation Sciences and Sustainable Technologies, 1(1)(2021), 21-37.

Received in final form on January 12, 2021

Abstract
Music classification is used by many companies such as Spotify and Soundcloud either as a recommendation system to their customers or only as a product such as Shazam. Determining the music genre is the first step in this direction. The major challenge in the music genre classification is to extract differentiating features from the audio data that could be fed into the model. Existing works classified the music by extracting hand crafted features (such as Mel-frequency cepstral coefficients, Spectral Centroid, and Chromal Features) and employing machine learning techniques (such as k Nearest Neighbours, Support Vector Machines, and Neural Networks). This work classifies music based on their genre using deep learning techniques, namely Convolutional Neural Network (CNN) and Recurrent Convolutional Neural Network (RCNN). We perform data augmentation on the audio data to generate synthetic data from the existing dataset such that the generalization capability of the model can be improved. Spectrograms are generated and are given as input to the network model. We used GTZAN (G. Tzanetakis and Cook[3]) music genre dataset for the experiments. The proposed approach provided an accuracy of 81.55 percent for four layers 2D CNN model and 82.05 percent accuracy for RCNN model by performing data augmentation; and the model provided an accuracy of 69.49 percent for four layers 2D CNN model, and 77.92 percent accuracy for RCNN model without data augmentation. Experimental results show the viability of the proposed model. Also, we examined the influence of each augmentation technique on the classification accuracy of the model for each class. We have observed that the performance of the classifier is influenced differently by each augmentation technique.


Keywords
Music Genre, Data Augmentation, Spectrograms, Recurrent Convolutional Neural Networks


Cite This Article
Macharla Vaibhavi and P. Radha Krishna, Music Genre Classification using Neural Networks with Data Augmentation , J. Innovation Sciences and Sustainable Technologies, 1(1) (2021), 21-37. https://doie.org/10.0608/JISST.2022772635


    432    37    Download