Unsupervised Music Genre Clustering Using Contrastive Learning

Disciplines

Artificial Intelligence and Robotics

Abstract (300 words maximum)

Music genre classification is a challenging task, especially in the absence of labeled data. In this project, we leverage unsupervised contrastive learning to cluster music tracks based on their underlying audio features. By applying SimCLR-based feature extraction on the GTZAN dataset, we demonstrate that contrastive learning can capture meaningful representations of different genres. Our approach does not require labeled training data and provides genre similarity clustering based on learned embeddings. Results indicate that the model effectively groups tracks into clusters that align well with traditional genre labels, suggesting contrastive learning as a powerful tool for unsupervised audio analysis.

Academic department under which the project should be listed

CCSE - Computer Science

Primary Investigator (PI) Name

Md Abdullah Al Hafiz Khan

This document is currently not available here.

Share

COinS
 

Unsupervised Music Genre Clustering Using Contrastive Learning

Music genre classification is a challenging task, especially in the absence of labeled data. In this project, we leverage unsupervised contrastive learning to cluster music tracks based on their underlying audio features. By applying SimCLR-based feature extraction on the GTZAN dataset, we demonstrate that contrastive learning can capture meaningful representations of different genres. Our approach does not require labeled training data and provides genre similarity clustering based on learned embeddings. Results indicate that the model effectively groups tracks into clusters that align well with traditional genre labels, suggesting contrastive learning as a powerful tool for unsupervised audio analysis.