# Music genre classification with modified residual learning and dual neural network

## TL;DR

This paper introduces a new deep learning model for classifying music genres using modified residual learning and dual neural networks, achieving strong accuracy on standard datasets.

## Contribution

The novel contribution is a hybrid CNN architecture with modified residual learning for improved music genre classification.

## Key findings

- The model achieved 87.80% accuracy on the GTZAN dataset.
- It obtained 68.50% accuracy on the FMA dataset.
- Performance is comparable to state-of-the-art models.

## Abstract

Music Genre is an abstract property of music that can identify shared traditions and conventions. In the recent past, music genre classification has shown a significant role in MIR that has attracted the research community to draw attention all around the world. The subjective aspect of the genre makes it challenging to define, as it relies on listeners’ interpretation. Deep Neural architectures can be used to address the efficiency and accuracy issues of traditional music systems. This paper proposes an approach to improve the music genre classification tasks with modified residual learning and hybrid convolutional neural networks. This architecture exploits the Mel-Spectrograms as input, which compute the signals as perceived by humans. We use identical layers of CNN with different pooling techniques to give rich hidden information for classification. We trained our model with Mel-Spectrograms generated from music files and obtained an accuracy of 87.80% and 68.50% for the GTZAN and FMA datasets, respectively. Our results show that the performance of the proposed model is also comparable with the other state-of-the-art models.