The cultural intelligence layer for music AI.
Documented artist agreements. Expert-annotated. Consent-logged at the source.
Hear the dataset.
Excerpts from the corpus, recorded in the studio at 96kHz / 24-bit. Full-resolution masters are available under licence.
The triple crisis.
Hindustani classical music runs on pitch, rhythm and lineage that general-purpose music AI was never trained to hear.
Cultural
Hindustani classical knowledge survives almost entirely through oral transmission from master to student, with no systematic archive built for computational study.
Creator
Before RaagaPay there was no commercially licensable, consent-logged corpus of Hindustani classical audio. The musicians who hold the tradition had no seat in the AI value chain.
AI bias
General-purpose music AI was not built around the vocabulary this music needs: continuous pitch and ornament, tala cycles of 7, 12 and 16 beats, and raag and gharana context. Trained on a fixed grid, a model loses what happens between the lines.
12 tones. Or 22 shrutis.
Today's AI hears a grid. A raag lives between the notes.
Each step is fixed, and nothing between the steps is represented.
Pitch moves through the space between the steps, where the ornament lives.
Synthesized tones for illustration only. Real corpus recordings are in the Listening Room above.
Existing models are trained almost entirely on 12-tone Western material. Shruti, raag and tala structure falls outside what they were built to represent.
Maestro-verified, consent-logged, research-grade data built for the microtonal reality of the music.
See howThe 80-point annotation framework.
We don't tag genre. We map what only trained musicians can hear.
Pitch anchor
Human-verified tonic and shruti labels on every track.
Stylistic lineage
Gharana lineage and training history documented for every performance.
Multi-stem masters
Vocals, sitar, tabla, harmonium, bansuri and sarangi, isolated and aligned.
Built on integrity. Not scraping.
Musicians are paid upfront and on every use. You get commissioned, consent-logged recordings.
Methodological rigour
Every data point verified by trained musicians. Every process documented for reproducibility.
96kHz / 24-bit
Studio-grade fidelity. Every microtonal nuance preserved.
80-point annotation framework
Raag, taal, gharana, laya, shruti, all annotated per track.
FAIR compliant
Findable, Accessible, Interoperable, Reusable.
100% artist consent
Documented consent, attribution and royalty agreements.
Fair trade AI
Upfront pay
Paid immediately on recording.
Lifetime royalties
Royalties on every use.
Commissioned recordings
New studio recordings, commissioned directly from each artist, with every performer's consent logged.
Cultural credit
Lineage attribution, always.
Choose your licence.
Three structured tiers built for the realities of academic study, commercial AI development, and enterprise procurement.
Research pilot licence
A starter corpus for non-commercial study, citation and publication.
Commercial development licence
The full production corpus, licensed for commercial model training.
Enterprise licence
Multi-year strategic access, co-branding and direct artist liaison.
Four steps. One loop.
From the studio to the model to the maestro's pocket. A loop that pays back.
-
01
Record
A maestro performs in the studio, on their terms.
-
02
Preserve
Every nuance archived at 96kHz against the 80-point annotation framework.
-
03
Train AI
Partners train on commissioned, consent-logged recordings.
-
04
Share value
Royalties flow to the maestro on every use.
Scaling the corpus.
From a verified pilot corpus to production-scale supply. Hindustani first, Carnatic in Phase 2.
Recorded, annotated end to end, and 100% consent-logged.
135 studio tracks, the 80-point annotation framework applied end to end, ethical protocols live.
Corpus expansion under way, the consent-signed artist roster growing, and the Carnatic recording programme in preparation.
Who's behind RaagaPay
Debjit Mitra
Musician and entrepreneur at the intersection of Indian classical music and AI.
Background spanning Spotify, the BBC, Zee TV and Ableton.
Sets product, strategy and the artist-first model.
The industry is listening.
Independent coverage of RaagaPay across music-industry press, legal analysis and AI commentary.
New Indian startup RaagaPay wants to fix AI's Hindustani classical music problem
Feature interview with founder Debjit Mitra on closing the Hindustani training-data gap.
Keeping Ownership Authentic: AI and Control in the Modern Matrix of Music
Cites RaagaPay's royalty-based dataset model among consent-forward approaches to AI training.
Global Cultures Are Expanding AI Music
A survey of builders teaching AI non-Western musical systems, with RaagaPay named for its Hindustani corpus and per-performance metadata.
Delhi startup RaagaPay builds the first ethical AI dataset for Hindustani classical music with lifetime artist royalties
On the 80-point annotation framework and the lifetime artist royalty model.
Frame: AI Music Isn't Taking Work. It's Moving It.
AI music culture and analysis on where the work, and the value, in music is actually moving.
Writing about RaagaPay? For interviews, comment or corpus documentation, write to debjit@raagapay.in.
Questions, answered.
What AI teams, researchers and procurement ask us most, before the rate card.
RaagaPay is the cultural intelligence layer for music AI. We record, annotate and license the first commercially licensable, consent-logged corpus of Hindustani classical audio, built for enterprise AI training and evaluation, with artist consent and royalty terms attached to every track.
Traditional raag performance, plus performances inspired by raag-based film songs. The Phase 1 corpus holds 135 studio tracks of that repertoire, recorded at 96kHz/24-bit as multi-stem masters across six instrument and vocal categories: tabla, bansuri, harmonium, sitar, vocals and sarangi. Every track is a new studio recording commissioned directly from the artist.
Every track is annotated by trained musicians across five categories: performance parameters, performer information, cultural context, technical analysis, and preservation and rights. The field-level schema is shared with partners under agreement.
The corpus is 100% consent-logged at source. Every artist signs documented consent that explicitly authorises AI training use, is paid upfront for the recording session, and earns royalties on every subsequent use, with lineage attribution preserved in the metadata.
Three tiers: a research pilot licence for non-commercial academic study from $2,500 for twelve months, a commercial development licence for AI model training on annual terms, and enterprise agreements for multi-year strategic access. Commercial terms are provided through the rate card.
Yes. Under a signed NDA and dataset evaluation agreement, we provide an evaluation pack of six full-length tracks with their complete annotation records, on a two-week evaluation licence, typically delivered within 48 hours.
Not yet. The current corpus covers Hindustani classical music only. Carnatic recording is planned for Phase 2 of the roadmap, and enterprise partners receive priority access as it comes online.
RaagaPay, operated by The Sonic Story Private Limited in New Delhi, India, is the contracting party of record on every licence. Artists retain their documented rights and royalty entitlements under their signed agreements, and consent records travel with each track.
Every track is a new studio recording commissioned directly from the artist, and nothing is licensed in from a film catalogue. The film-song-inspired performances take a raag-based film song as their starting point. A licensee receives a new Hindustani classical performance, not a film master.
A first internal evaluation has been run across raag identification, tala recognition, pitch and ornament tracking, and lineage retrieval. We publish figures alongside the method rather than ahead of it, so results are released once the protocol is documented and comparison models have been scored on the same tasks. Licensees can run the evaluation themselves under a dataset evaluation agreement.
A shruti is one of 22 microtonal pitch positions used in Indian classical music. Western tuning divides the octave into 12 semitones; Indian classical music works with 22 finer steps, and the music lives in the space between them.
A raag is a melodic framework, not just a scale. It sets rules for which notes to use, how to approach and leave them, and the mood they create. Many raags are tied to a time of day or a season.
A gharana is a stylistic lineage passed from master to student over generations, similar to a house style, with its own signature techniques and repertoire.
A taal is the rhythmic cycle that structures a performance, commonly 6 to 16 beats, kept and elaborated by the tabla.
Build with the tradition,
not around it.
The first commercially licensable, consent-logged corpus of Hindustani classical audio. License it for training, or apply for research access.