The move from physical to digital distribution multiplied the volume of music uploaded to streaming platforms, but the tooling for the descriptive and administrative data attached to each recording never scaled with it. When a track reaches Spotify, Apple Music or YouTube Music it carries tags that were usually typed by hand — incomplete, inconsistent, or wrong.
Consequences:
- Discovery loss — recommendation and editorial systems can't surface a track that has no genre, mood, BPM or key.
- Royalty leakage — PROs (ASCAP, BMI, GEMA, ZAiKS…) can't allocate money for recordings they can't match; it sits in "black box" accounts.
- Manual cleanup cost — distributors and labels pay people to normalise metadata before delivery.
One curator's "Deep House" is another's "Tech House." The industry has a transport standard (DDEX) but no agreed way to decide what the data should say.
It analyses the audio signal itself and derives the descriptive data from measurable features rather than opinion:
- objective DSP metrics (tempo, key, loudness, spectral balance, dynamics);
- a multi-model AI classification that must reach consensus across independent models before a tag is accepted, each tag carrying a confidence score the user can override;
- identification against public databases (AcoustID, MusicBrainz, ACRCloud);
- a SHA-256 Authenticity DNA hash of the decoded audio that ties the metadata to that exact recording regardless of file format;
- a signed Certificate of Authenticity, pinned to IPFS, publicly verifiable.
| Segment | Why they use it |
|---|---|
| Independent artists & producers | Get the BPM/key/genre a distributor (DistroKid, CD Baby, TuneCore) expects, without guessing. |
| Labels & publishers | Batch-audit a back catalogue for missing descriptors so every asset stays discoverable and royalty-eligible. |
| Distributors / DSPs | Verify incoming metadata in the ingestion pipeline; fewer manual reviews, fewer rejected deliveries. |
| Sync & production-music libraries | Granular mood / energy / instrumentation tagging so supervisors can actually search the catalogue. |
- vs Shazam / SoundHound — those match a recording to a known fingerprint database ("who is this"). Metadata Engine describes structure ("what is this") and works on unreleased material.
- vs Gracenote — a centrally curated database vs a signal-first approach whose output (the Authenticity DNA + certificate) can live locally or on IPFS.
- vs generative audio models (MusicLM, AudioCraft) — those create audio; the moat here is standardisation and certification, not the analysis alone.
One-time credit packs (no subscription). 1 credit = 1 full analysis; exports after an analysis cost nothing extra. New accounts get 3 free credits.
| Pack | Credits | Price (USD) |
|---|---|---|
| Starter | 10 | $9 |
| Producer | 50 | $35 |
| Label | 150 | $89 |
| Studio | 400 | $199 |
Payments run through Stripe. HardbanRecords Lab is the seller of record (see LEGAL-OVERVIEW.md). Promo/referral redeem codes grant credits directly.
Implementation detail for operators is in OPERATIONS.md; the billing flow is in ARCHITECTURE.md.
- Near term — tighten the AI genre taxonomy, faster analysis, redeem/ referral growth loop.
- Mid term — a partner/enterprise API with async batch + webhooks; embedding the Authenticity DNA into the audio bitstream (watermarking).
- Long term — richer generative descriptions, a verifiable metadata ledger, PRO/CMO partnerships around the Authenticity DNA standard.
The 2026 whitepaper draft (previously
white peper.txt) described some capabilities aspirationally (Kubernetes worker clusters, trained CNN/ Transformer classifiers, a WooCommerce credit system). The shipping system is simpler and is what ARCHITECTURE.md documents.