Key Highlights
- RaagaPay uses 80 metadata parameters to classify Hindustani classical music, compared to 12-15 in Western datasets
- Artists receive lifetime royalties each time their recordings are licensed for AI training
- The startup targets 1,000 hours of curated recordings from 100+ performers across 50 gharanas by 2028
AI music tools fail Non-Western traditions
Ask Suno AI for a raga Yaman and you get a Western approximation. That failure rate reaches 40% hallucination when AI models attempt to generate Hindustani classical music, according to RaagaPay founder Debjit Mitra.
The Delhi-based startup launched in October 2025 to fix this structural gap. Mitra, a sound engineer and composer who has worked with Spotify, BBC, and Zee TV, built a dataset requiring roughly 80 metadata parameters. Western systems use only 12-15.
80 Parameters Capture What Western Datasets Miss
The missing parameters in current AI training data include:
- Raga classification (melodic frameworks with specific ascent/descent rules)
- Taal (complex rhythmic cycles like Teentaal’s 16 beats or Jhaptaal’s 10 beats)
- Rasa (a codified system of 9 classical emotions)
- Season and time-of-day associations
- Gharana affiliation (traditional school/lineage affecting how the same raga sounds)
“We want to solve that problem, and beyond that preserve our culture and heritage,” Mitra told Music Ally.
Lifetime royalties model addresses consent gap
RaagaPay completed Phase 1 in December 2025 with 10 cumulative hours across sitar, harmonium, tabla, bansuri, sarangi, and vocals. The recordings capture sargams (skeletal melodic notes) stripped of original lyrics and performer-specific ornamentation. This isolates the core raga structure without exposing performer IP.
Artists receive lifetime royalties each time their recordings are licensed for AI training. This consent-based approach mirrors what the Fairly Trained certification program promotes and what rights holders want from AI companies facing RIAA copyright lawsuits.
Global template for regional music AI datasets
This is the first startup targeting the training data gap for a non-Western classical tradition. The same problem exists for Arabic maqam, Turkish makam, Persian dastgah, Carnatic music, and West African griot traditions.
The parallel to speech recognition bias is instructive. Early systems performed poorly on non-English accents and languages. AI music models are at that same early stage, trained almost entirely on Western tonal music.
South Korean startup Pozalabs took a similar ethical approach by building over 1 million original training samples with in-house composers. RaagaPay extends this model to regional traditions.
Producers should treat AI tools as unreliable for Ragas
If you work with clients in non-Western music traditions, treat current AI music tools as unreliable for raga, maqam, makam, or similar frameworks. Do not use them to generate reference tracks or demos without manual correction from a trained musician.
FAQ:
Q: Why do AI music generators fail at Indian classical music?
Hindustani classical music requires approximately 80 metadata parameters to describe accurately, including raga classification, taal, rasa, and gharana affiliation. Western AI training datasets use only 12-15 parameters and cannot capture these structural elements.
Q: What is RaagaPay and how does it pay artists?
RaagaPay is a Delhi-based startup building ethical AI training datasets for Hindustani classical music. Artists receive lifetime royalties each time their recordings are licensed for AI training, creating a consent-based compensation model.
Q: Can Suno generate authentic Indian classical music?
No. According to RaagaPay founder Debjit Mitra, AI models trained on scraped data hallucinate as much as 40% of the time when generating Indian classical music. Current tools produce Western approximations rather than authentic ragas.
Q: Does the AI training data problem affect other non-Western music traditions?
Yes. The same structural gap exists for Arabic maqam, Turkish makam, Persian dastgah, Carnatic (South Indian classical), and West African griot music. RaagaPay serves as a proof-of-concept for ethical regional music AI datasets globally.
