Small
For a handful of tracks or occasional checks.
€9one-time
10 credits · €0.90 per scan
- 10 completed master scans
- Full master report every time
- Reports saved in My Scans
- Purchased credits valid for 12 months
How Trace works
Evidence, not guesswork.
01 · Upload
Your master goes in.
Upload WAV, MP3 or FLAC. Trace reads the audio structure and checks available provenance and metadata.
02 · Analyze
Trace looks inside the audio.
Forensic audio signals are compared and mapped across the track. Trace builds a spectrogram and identifies localized evidence regions.
03 · Report
See where the evidence is.
Your report combines provenance, master-level signals and timeline evidence. Click the waveform to hear the regions Trace is referring to.
04 · Go deeper
Deep Forensics is next.
Where deeper examination is useful, stem-level analysis can separately examine vocals, drums, bass and other elements.
What the result means
Trace is a forensic screening tool. It reports characteristics and evidence found in the submitted audio; it is not proof of origin. Results remain available in My Scans.
Catalyse Trace · Master scans
Start free.
Scan more when you need.
Explore AI-related evidence in your music. Buy a credit pack when you need more scans, with the full master report included every time.
- One-time payments
- No subscription
- No credit card to try
Simple pricing
Choose how much you want to scan
Same complete master report. More credits when you need them.
Standard
For regular checks across your releases.
€19one-time
25 credits · €0.76 per scan
- 25 completed master scans
- Full master report every time
- Reports saved in My Scans
- Purchased credits valid for 12 months
Large
For more tracks and a lower cost per scan.
€39one-time
60 credits · €0.65 per scan
- 60 completed master scans
- Full master report every time
- Reports saved in My Scans
- Purchased credits valid for 12 months
Launch pricing. Credit purchases are not enabled yet. You can try Trace with your two free introductory credits today. Applicable tax details will be shown at checkout when purchases open.
Full results, every time
See the findings and evidence coverage from your master scan in one report.
Only completed scans count
Failed processing does not use a credit. An inconclusive completed report still counts as a scan.
Your reports stay accessible
Reopen a saved report in My Scans without spending another credit.
A few useful answers
What does one credit cover?
One completed master scan, including its full current report. These packs cover master scans only.
Do the two free credits reset each month?
No. They are an introductory allowance for new verified accounts. When you need more scans, choose a credit pack once purchases open.
How long will purchased credits last?
Purchased credits will be valid for 12 months from purchase. There is no subscription and no automatic renewal.
Can I reopen a report for free?
Yes. Opening a saved report does not cost credits. Running a new analysis uses one credit.
Does a report prove that a track is AI-generated?
No. Trace reports available provenance and AI-related characteristics, with evidence coverage and limitations. Findings are indicators, not proof of origin.
Simple packs. Complete master reports. No subscription.
Uploading your audio
Saving and preparing your upload
Checking provenance & metadata
Reading audio file
Preparing audio for analysis
Comparing forensic evidence patterns
Building the spectrogram
Mapping evidence across the track
Calculating evidence regions
Preparing your report
Trace account
Sign in to scan
New accounts receive the current included-credit allowance. Your reports stay in My Scans.
We’ll email you a secure sign-in link. No password needed.
My Scans
Your tracks. Your findings.
Product settings
Users
Mistake Bank
← Back to upload
AI characteristics detected.
This is a forensic screening result, not proof of origin.
Track
Master audio
Evidence strength
No significant evidence
0%Evidence coverage
Production type
Unresolved
Insufficient evidence for production-type attribution.
▤
Provenance & metadataChecked
✓≋
Master-level audio signalsIndicators found
!☷
Timeline analysisRegions identified
!Waveform & RegionsSpectrogramTechnical detailsRegion list
0:00 / 0:00No evidenceTap waveform to jump
No evidenceWeakModerateStrong
Spectrogram (master)
4–10 kHz frequency view with detected regions.
10 kHz8 kHz6 kHz4 kHz
Detected regions
Click a region to listen.
| # | Start | End | Duration | Evidence |
|---|
⚙ Technical detailsEngine · sample rate · reference pack ⌄
What we found
Already in My Scans
This file has already been analysed
Only an identical file in your own account is matched. An edited export is treated as a new file.
Explainer
AI audio watermarks and detection: what is the difference?
An inserted watermark and a forensic audio finding are different clues. Neither a missing marker nor a reassuring scan establishes human authorship.
Catalyse TRACE Research · 2026-10-03 · 2 min read
ai musicprovenancewatermarks
A tool says it found an AI watermark. Another reports suspicious audio measurements. Those statements describe different checks.
A watermark is deliberately added to eligible content. Passive forensic analysis examines characteristics of the submitted audio without requiring that a particular marker was inserted during generation.
Watermarks depend on the system
Google DeepMind describes SynthID as a watermarking system for AI-generated media, including audio in supported products such as Lyria and NotebookLM. Its audio watermarks are designed to remain detectable through common modifications. That description is not a guarantee for every transformation or a marker shared by every music generator.
AudioSeal, an ICML 2024 research paper by Roman and colleagues, studies localized watermarking for generated speech. Speech research is useful context, but it should not be presented as validation for all generated music.
Absence is a limited observation
If a check finds no supported watermark, several questions remain. Was a marker inserted in the first place? Is the provider supported by that check? What transformations happened to the file?
The absence of a detected marker does not answer those questions or establish that the music was made entirely by humans.
Presence also needs scope
A reported marker should be described with its source and verification limits. It may offer a clue about participating tools; it does not automatically explain every instrument and creative decision in a mixed production.
Ordinary metadata, signed provenance credentials and embedded audio watermarks are also distinct mechanisms. When reading a report, keep the name and scope of each finding visible.
Read the layers together
Ask which checks produced information and which did not. Then consider those findings alongside the audio measurements and available production records.
For TRACE, this is an interpretation guide. It does not claim that TRACE implements SynthID or AudioSeal verification. A provider-specific capability should only be described as supported when the service explicitly documents it.
A useful report explains what was found and how far that evidence reaches. It should leave room for the questions that a single marker cannot resolve.
Sources & further reading
- deepmind.google — https://deepmind.google/models/synthid/
- arxiv.org — https://arxiv.org/abs/2401.17264
- trace.catalyse.no — https://trace.catalyse.no/research/audio-metadata-provenance-before-analysis
An explainer is educational guidance. Test notes describe documented experiments; neither format certifies an individual track.
What’s inside your track?
Explore the signals, then read them in context.
Scan your trackHow TRACE worksPricingKeep exploring
Explainer
What your audio file reveals before we hear it
Format, file structure and provenance clues can add context to an audio scan. Missing metadata is not evidence of human authorship.
2026-10-03 · 3 min read
provenanceaudio metadataai music
Explainer
Can audio processing remove an AI watermark?
Some processing can weaken detection in specific systems. That is not a universal watermark-removal recipe—or proof that a track is human-made.
2026-10-03 · 3 min read
watermarksaudio formatsmusic production
Explainer
Can you hide an AI watermark inside a hybrid track?
Adding human instruments changes a mix—not its history. Learn why audible masking, watermark detection and AI audio evidence are different questions.
2026-10-03 · 3 min read
watermarkshybrid musicmusic productionProduct information · Pre-launch draft
Terms & Conditions
TRACE is currently in testing. This page outlines the current service. Final purchase terms and privacy information will be published before payments and public launch.
Understanding your report
TRACE screens submitted audio for provenance, metadata and measurements associated with AI-generated material. A result is not proof of who created a track. Ordinary production and processing may explain suspicious measurements, and a scan may miss AI-generated material.
Your uploads and account
Only upload audio you have permission to submit. Saved scans are accessed through your signed-in account. An identical upload in your own account can open its saved report for free; a new analysis requires confirmation of its displayed credit cost.
Credits and pricing
Credit purchases are not yet enabled. Launch packs are €9 for 10 credits, €19 for 25 credits and €39 for 60 credits. One credit covers one completed master scan; failed processing does not consume a credit. Purchased credits will be valid for 12 months from purchase, with no subscription or automatic renewal. Final purchase and refund terms will be published before checkout opens.
Audio storage and privacy
Audio is stored to support saved-report playback and is automatically deleted 30 days after a completed scan. The report remains in My Scans after the audio expires. You can delete a scan earlier at any time; deleting a scan removes its saved report, stored audio and playback copy, and revokes its report invitations. A fuller privacy policy will be published before public launch.
Optional Google Analytics
If you accept analytics, TRACE uses Google Analytics to understand visits and how features are used. Google receives usage events and browser/device information and processes connection information such as your IP address to provide the service. Analytics may store cookies in your browser. We do not send your audio, filenames, email address, report findings, private report IDs or invitation tokens in analytics events. Advertising features are disabled. Rejecting analytics does not affect scanning or sharing. You can change your choice at any time using the Analytics preferences button below. Essential sign-in and security storage remains necessary to use your account. Read Google’s privacy policy for information about Google’s processing.
Contact
For product enquiries, visit Catalyse.