In your upload or QC flow
Call SEAM as tracks arrive, and keep each verdict next to the release in your review queue.
- Paper Moonswaiting for reviewAI 52%
- Slow Harbourwaiting for reviewAI 4%
Built by The InterSpace Group
Sonic Evidence of AI in Music
An API for AI music detection. Send a track from your upload or QC pipeline and get back a verdict on how likely it is to be AI-made. Your team makes the call.
Try the console
Pick a sample track and SEAM scans it again. These are sample tracks showing the report format, not real songs or customers.
Sample tracks
Made-up tracks. Pick one to scan it again.
Verdict
AI line: 80%
Scanning
Account check
Checked once the windows are read.
Advisory: your team decides.
API
SEAM is a REST API, JSON in and out, with keys issued per account. Plug it into your own upload or QC pipeline, and every track you send comes back with a verdict.
From your upload flow, your QC queue or a back catalogue.
Windows from across the track, each scored, then weighed together.
likely_ai, unsure or likely_human, with an AI likelihood from 0 to 1.
{ "verdict": "likely_ai", "ai_likelihood": 0.97, "model": "v3b", "scanned_at": "2026-11-03T09:14:07Z" }
{ "verdict": "unsure", "ai_likelihood": 0.52, "model": "v3b", "scanned_at": "2026-11-03T09:15:42Z" }
{ "verdict": "likely_human", "ai_likelihood": 0.04, "model": "v3b", "scanned_at": "2026-11-03T09:16:18Z" }
An example with sample values. Four fields: verdict, ai_likelihood, model and scanned_at.
Call SEAM as tracks arrive, and keep each verdict next to the release in your review queue.
Send a back catalogue or a new intake, track by track. Every track gets its own verdict, so your team can start with the ones that need a listen.
API access starts once we agree a quote. Your first 3,000 tracks are scanned free.
Get a quoteSEAM runs on The InterSpace Group's own servers.
Under the hood
Sample values from the made-up tracks above.
What you get
Every track comes back with an AI likelihood and a verdict band, worked out from the sound itself.
SEAM reads windows from across the track, so an intro or a fade-out doesn't decide. A detector scores each window for the traces AI generators leave in the sound, and SEAM weighs those scores with statistics across the windows, like how well they agree.
Sample track. Each window is scored on its own.
Titles, credits and AI disclosures can say anything. SEAM scores the audio itself, so a missing disclosure doesn't change the verdict. It's built to catch music from the major AI music generators.
Sample track.
One track can be hard to call. When most of an uploader's other tracks look AI-made, SEAM lowers the line for that account's tracks, from 80% to as low as 31%. It can raise a score, never lower one, and it helps most with accounts that have plenty of uploads.
Sample account: a 66% track now reads 84%.
Every verdict is advisory. SEAM never rejects, holds or approves a track on its own. Your team decides what each verdict leads to, under your own policy.
Likely AI. Strong signs of a generator.
Unsure. A reviewer listens and decides.
Likely human. No strong sign of a generator.
Who it's for
Check demos and deliveries before you sign or release them.
Score every upload while it waits in QC.
Scan the catalogues your clients send before they reach stores.
Look twice at AI-likely deliveries before they go live.
Check the works and catalogues you take on.
Know what you are licensing before your clients ask.
Screen entries before they reach the judges.
Check submissions before they go on air.
Questions
No. SEAM gives an advisory verdict, and your team decides what happens next, under your own policy.
Through the SEAM API: a REST API, JSON in and out, with keys issued per account. API access starts once we agree a quote.
A verdict (likely_ai, unsure or likely_human), an AI likelihood from 0 to 1, the model version and the time of the scan.
No. SEAM scores the sound. Your reviewers can weigh the metadata next to the verdict.
It's built to catch music from the major AI music generators, by the traces they leave in the sound.
Which account uploaded each track. It looks at that account's other uploads, and it can only raise a score, never lower one.
Every team gets its own quote, based on the use case and how many tracks you scan. Tell us about yours and we'll come back with a quote. Your first 3,000 tracks are scanned free.
Not on this page: the console uses made-up sample tracks to show the format.
On The InterSpace Group's own servers.
Get a quote
We'll come back with a quote. Every team gets its own, based on the use case and how many tracks you scan.
3,000Free scans for your first 3,000 tracks.
Our team will be in touch with a quote.