Indian Classical Music Synthesis

Download paper here

Several studies in the fields of Artificial Intelligence and Natural Language Processing have been conducted on Music Synthesis. However, due to the limited availability of structured datasets, the field of Indian Classical Music remains unexplored. Additionally, the considerable influence of western music in the past decade has adversely affected the market and demand for Indian Classical Music. In this work, we propose a model to generate music for Indian Classical Music, specifically Carnatic Music, by leveraging the structured nature of Indian Carnatic Music. We build a dataset of Classical Indian Music with paired lyrics and melody to map melody with notes to extract features. Generative Adversarial Networks (GANs) are proven to be very effective for music generation in several research works. We experiment with GANs and Auto Encoder (Variational AE and Conditional VAE) on classical lyrics. The curated dataset shall also be helpful for further research in the domain of Indian Classical Music.

Recommended citation: Gaurav Viramgami*, Hitarth Gandhi*, Hrushti Naik*, Nipun Mahajan*, Praveen Venkatesh, Shivam Sahni, Mayank Singh. "Indian Classical Music Synthesis." ACM India 5th Joint International Conference on Data Science and Management of Data [CODS-COMAD 2022].
Download Paper