Delhi-based startup RaagaPay claims to be “India’s first ethical Indian classical music dataset for AI”, specifically Hindustani classical music. “When you ask a Suno to give [you a song in the] raga Yaman, it will give you a Western approximation of the raga,” says founder, sound engineer and composer Debjit Mitra.
He launched RaagaPay in October 2025 to fix this. “We want to solve that problem, and beyond that preserve our culture and heritage,” says Mitra, who has worked with Spotify, the BBC and Zee TV and runs the sonic branding company The Sound Story. To this end, he has been recording Indian classical musicians across 50 gharanas or schools of music.
The first phase, which comprised a cumulative ten hours of sitar, harmonium, tabla, bansuri (Indian bamboo flute), sarangi and vocal music, was completed in December. His aim is to lay down 1,000 hours of recordings by over a 100 performers by 2028. He told us that the artists will be paid lifetime royalties for their work, which they will earn every time a track of theirs is licensed for Al training or other uses.
Mitra says that while most “Western metadata schema [covers] about 12 to 15 to parameters, like what tempo a song is in, what key it is in,” RaagaPay’s tally is around 80 and includes a whole lot of genre-specific parameters such as the classification of the raga, the taal (rhythm), its emotional framework (rasa), the season and time it’s associated with, and the gharana to which it belongs.
Interestingly, 50% of the recordings will be of Hindi film music, and the metadata includes information such as the use of a raga in a particular Bollywood song, for e.g. in a romantic track composed by S.D. Burman or R.D. Burman. “At the end of the day, all Bollywood music is based on ragas,” says Mitra whose team includes an ethnomusicologist, a human annotator and a lawyer.
When asked how he was able to record existing compositions, he explained that they were specifically capturing the sargams or notes of the songs that form its skeletal melodic framework (sa, re, ga, ma…) “stripped of the original lyrics, specific orchestration/arrangement, production elements and the performer’s unique interpretation”.
“This falls under un-copyrightable musical ideas rather than protectable expression,” says Mitra who added that copyright “protects the specific recorded performance and arrangement (the sound recording and musical work), not the underlying melodic idea or scale pattern. If recorded as neutral, instructional renditions not mimicking the original arrangement, this is transformative and educational, which strengthens fair use/fair dealing arguments under Section 52 (1)(a) of the Indian Copyright Act.”
The next step for the company, he says, is to enter into a strategic partnership with academic and research institutions, and then gradually tie up with AI companies.“We’re telling these companies, your model hallucinates as much as 40% with the data you’ve scraped, take our data and teach it. Your convergence rates will be better.”
‘ The preceding article may include information circulated by third parties ’
‘ Some details of this article were extracted from the following source musically.com ’
