Indian Startup RaagaPay Building AI Dataset for Hindustani Classical Music

A Delhi-based startup called RaagaPay is working to create a dedicated dataset for Hindustani classical music to train artificial intelligence systems more accurately.

The company was founded in October 2025 by sound engineer and composer Debjit Mitra. The project aims to address a common problem with AI music tools that often struggle to accurately reproduce the structure and style of Indian classical ragas.

According to Mitra, when many AI tools are asked to generate music in a raga such as Yaman, the result often sounds closer to Western music rather than an authentic Hindustani interpretation. RaagaPay’s dataset is intended to provide more accurate training material for AI systems.

To build the dataset, the company has begun recording musicians from around 50 gharanas, or traditional schools of Hindustani classical music. The recordings include performances featuring sitar, harmonium, tabla, bansuri, sarangi, and vocal music.

The first phase of the project was completed in December 2025 and includes about ten hours of recordings. RaagaPay plans to expand the archive to around 1,000 hours of recordings from more than 100 musicians by 2028.

The company says artists contributing to the project will receive lifetime royalties whenever their recordings are licensed for AI training or other purposes.

RaagaPay is also developing a detailed metadata system for the recordings. While many Western music databases track around 12 to 15 parameters, the RaagaPay dataset is expected to include around 80 parameters. These will cover aspects such as the raga, taal (rhythmic cycle), emotional mood or rasa, time, and seasonal association of the raga, and the gharana tradition.

The project will also document the connection between ragas and Hindi film music, since many Bollywood compositions are based on classical ragas. Around 50 percent of the recordings will reference film music, including examples connected to composers such as S.D. Burman and R.D. Burman.

To avoid copyright issues, the recordings focus on the sargam framework of compositions—note sequences such as sa, re, ga, ma. These recordings exclude original lyrics, orchestration, and stylistic elements from specific performances.

According to Mitra, the aim is to document the underlying melodic structure of ragas rather than reproduce copyrighted recordings.

RaagaPay plans to collaborate with academic and research institutions in the next stage of the project and later work with AI companies interested in improving how their models understand and generate Indian classical music.