AI Data Broker for Music & Audio: License Music for AI Training
AI music is moving to licensed models. Your catalogue could be part of it.
As an AI data broker for music and audio, Signal & Form helps Canadian labels, publishers, and studios license catalogues to AI companies on their own terms. Labels, publishers, and studios can now license music for AI training on their own terms, as AI music companies shift from scraped data to licensed catalogues. Stems, production music, sound libraries, and spoken-word archives are all in demand. We help Canadian rights holders work out what their audio is worth, untangle the splits and consents, and reach the right buyers.
Your music may already be training someone else’s model
If you run a label, a publishing company, a studio, or a podcast network, you have probably wondered whether AI music generators trained on your work without asking. You are not alone. In September 2026, SOCAN sued the AI music generator Suno in Federal Court, alleging it trained on its members' music without permission.
At the same time, there is a pile of audio that earns nothing: multitrack sessions on old drives, library cues that stopped placing, sound effects recorded for one project, and years of podcast episodes. The obvious question is whether you can get paid for that archive, legally and without giving away your artists' voices.
Increasingly, yes. But music has more owners per file than almost any other kind of data, so the work is in proving who owns what and who has consented.
What Sells
Data in music & audio that AI buyers want
Stems and multitrack sessions
Isolated vocals, drums, bass, and instrument tracks. Separated parts are hard to get from finished recordings and are useful for teaching models how music is built.
Production and library music
Catalogues with consistent metadata such as tempo, key, mood, genre, and instrumentation. Well-tagged libraries are easier to train on and easier to value.
Sound effects and field recordings
Foley, ambiences, and location recordings, especially original material that has not been widely published, with clear descriptions of what each file contains.
Podcasts and spoken-word archives
Long-form conversation with transcripts. Spoken audio is useful to speech and voice developers, but every voice in it is personal information, and voiceprints are treated as sensitive biometric data.
Who is licensing music, and on what terms
The clearest signal comes from the major labels. In October 2025, Universal Music Group settled its lawsuit against Udio and agreed to build a new platform trained on authorized and licensed music, with UMG artists and songwriters able to opt in and be paid for training and outputs. In November 2025, Warner Music Group settled with Suno, which said it would launch licensed models in 2026, with an opt-in for artists' names, likenesses, voices, and compositions.
Independent rights holders have deals too. When ElevenLabs launched its music model in August 2025, it announced agreements with Merlin, which represents independent labels and distributors, and the publisher Kobalt, letting their artists and songwriters take part on an opt-in basis.
Financial terms for these deals have not been published, so there is no public per-track price yet. What the deals do show is the shape buyers accept: opt-in consent, payment for training as well as output, and limits on using an artist's voice or likeness. Smaller catalogues are most attractive when they are well documented and fill a gap, such as a genre, a region, or a language.
Before You Sell
Rights and consent issues specific to music and audio
- Two copyrights in every song. The composition belongs to its songwriters and publishers. The recording belongs to its maker, usually the label or artist, under a separate right in the Copyright Act. A licence for training generally needs both sides, so owning masters alone may not be enough.
- Splits and co-writers. Songs with several writers, publishers, or administrators need every share accounted for. Missing split sheets are one of the most common gaps in a catalogue.
- Performers' rights. Canadian law gives performers their own rights in their performances. Session players, featured vocalists, and any applicable union agreements need to be checked before recordings are licensed.
- Samples and interpolations. Cleared samples were usually cleared for release, not for AI training. Tracks with uncleared or narrowly cleared samples may need to be left out.
- Voice is biometric. In October 2026, the Privacy Commissioner of Canada described voiceprints as sensitive biometric information and clarified when meaningful consent may be required. Podcast hosts and guests, voice actors, and vocalists should give explicit, written consent for this use. In Quebec, organizations must notify the Commission d'accès à l'information at least 60 days before creating a biometric database, and Law 25 requires a privacy impact assessment before personal information leaves the province.
- No voice clones without consent. Licences should restrict buyers from imitating an identifiable artist's voice or style unless that artist has agreed, the same opt-in approach the major label deals adopted.
General information, not legal advice. Have a lawyer review rights, consent, and any licensing contract.
The Canadian picture
Canada's music rights organizations are drawing a line on unlicensed training. SOCAN, which represents more than 185,000 music creators and publishers, filed its claim against Suno in the Federal Court of Canada in September 2026, seeking an injunction and damages. Music Publishers Canada is intervening in a Federal Court case about whether AI-assisted works can be protected by copyright, arguing that only a human can be an author.
Canada's Copyright Act has no text and data mining exception, and copyright collectives have asked Ottawa not to create one. The federal AI strategy released in June 2026 did not address copyright. For rights holders, that leaves a direct, documented licence as the clearest route to getting paid.
Canada also has catalogues that are hard to find elsewhere, from francophone music to regional scenes. Licensed well, with consent and clean splits, that becomes a form of data monetization that protects artists instead of bypassing them.
How It Works
From archive to licensing deal
We review what you hold, estimate a realistic value, work with your lawyer on rights and consent, document the dataset, and route it to the right buyers. You pay a success fee only if a deal closes. See the full process on our AI data licensing page.
FAQ
Questions about AI data broker for music and audio
Yes, if you control the rights needed and your artists agree. Independent labels and publishers have already been part of AI licensing deals, with artists opting in. Start with an inventory of what you own outright, what you share, and what consents you have, then estimate what the catalogue is realistically worth before preparing anything.
Generally, yes. A song has two copyrights: the composition, owned by its writers and publishers, and the sound recording, owned by its maker. Training on a recording uses both. Your lawyer should confirm what your agreements cover. This is general information, not legal advice.
They can be. Isolated parts are hard to get from finished mixes and show how a track is built, which makes them useful for training. Value still depends on volume, quality, metadata, and clean rights. No public per-track price exists yet, so any estimate should be treated as a range, not a quote.
Possibly, but voices change the picture. A voice is personal information, and Canada's Privacy Commissioner treats voiceprints as sensitive biometric information. You would need clear, written consent from hosts and guests for AI training, and episodes without it would usually be removed. Transcripts help, and so does a clean record of who appears in each episode.
Not if the contract prevents it. The major label deals use opt-in consent for names, likenesses, and voices, and you can require the same: training rights only, no voice clones, no prompts by artist name, and a defined term. Your lawyer reviews and signs off on every contract before any audio is delivered.
It shows Canadian rights holders are prepared to enforce their rights in Canadian courts, and it raises the value of music that buyers can license cleanly. It does not stop you licensing your own catalogue. If you think your work was used without permission, speak to your lawyer or your rights organization about that separately.
Related Reading
What AI Companies Pay for Training Data in 2026: Every Public Price, Sourced
How to Sell Your Data to AI Companies from Canada: Process, Buyers, and the Law
AI Voice, Video, and 3D Tools for Business: ElevenLabs, Seedance, Higgsfield, and Tripo
Other Industries
Sources: Canadian Lawyer: SOCAN sues Suno in Federal Court (Sep 2026) · Music Business Worldwide: Universal Music settles with Udio, licensed AI platform (Oct 2025) · Music Business Worldwide: Warner Music Group settles with Suno (Nov 2025) · Music Business Worldwide: Eleven Music launches with Merlin and Kobalt deals (Aug 2025) · Copyright Act, Part II: performers’ rights and sound recording makers (Justice Laws) · OPC: updated call recording guidance on voiceprints (Oct 2026) · Gowling WLG: Law 25 and biometric data in Quebec · Quebec private sector privacy act, s. 17 (LegisQuébec) · Music Publishers Canada: intervention in Federal Court AI and copyright case (Jul 2026) · Access Copyright and Copibec: 2026 pre-budget submission on GenAI and copyright (May 2026) · McCarthy Tétrault: Canada's AI strategy overlooks copyright
Sitting on music & audio data? Find out what it is worth.
Book a free call. Tell us what you hold, and we will give you an honest read on whether it is worth licensing.