Is this sound human or AI?
Upload a track or a voice: we examine it closely and you get a clear answer in 2 minutes. Suno, Udio, ElevenLabs, hybrid tracks…
Where in the track?
One bar every 2.5 s · the higher and more orange, the more it sounds like AIA statistical clue, not proof. For an important track, cross-check with careful listening and context.
What the tool analyses
Most detectors only look at the whole track. Yet a real voice over an AI-generated instrumental often slips under their radar. Here, every track is analysed from three angles.
The full track
AI probability passage by passage, every 2.5 seconds, to locate the suspicious parts in time.
The instrumental only
The vocals are removed by source separation, then the backing is analysed on its own. This is the most reliable clue.
The sung vocals only
The isolated vocals are also analysed and shown for information. On their own, they are not yet reliable enough to decide: they are not used for the verdict.
The spoken voice beta
Voice-overs, audiobooks, podcasts: detection of synthetic voices (ElevenLabs and others), passage by passage.
How it works
You upload a file
An mp3, a wav, an export from your DAW… up to 10 minutes analysed.
We separate and analyse
Vocals / instrumental separation, then models trained on nearly 100,000 real and generated tracks score each passage.
You get a clear verdict
A trend on a 4-step scale, the details per element, the timeline and a PDF report to download or pass on.
Who is it for?
Labels and distributors
Check a track you received before signing or distributing it.
Festivals, contests, radio
Screen applications and submissions when the rules exclude generated music.
Musicians and composers
Check a beat you bought or a sample, or prove the human part of your work.
Journalists and the curious
Find out whether that “new artist” taking off on the platforms really exists.
Frequently asked questions
Is the detector 100% reliable?
No, and no detector can be. The tool gives a probability. In November 2026, out of 545 real tracks it had never heard, 28 were classed “Likely AI” or higher, i.e. 5.1% false alarms: about one in twenty. The Suno tracks we tested score between 55 and 98%. Generators evolve fast: a very recent model or a reworked track can fool the detection. The details are on the Reliability page.
Which generators are detected?
The models are trained on tracks generated by Suno and Udio, the two most widely used song generators. Other generators may be caught if they leave similar traces, with no guarantee.
What happens to my file?
It is deleted as soon as the analysis ends. It is not listened to, kept or used to train a model. The PDF report can be downloaded for 7 days, then it is erased.
Can it detect a hybrid track, with a real voice over an AI instrumental?
Yes, that is what makes this tool different. The vocals and the instrumental are separated, then each is analysed on its own. On a hybrid track, the score of the full track can stay low while the instrumental alone clearly stands out.
Can it detect an AI-generated spoken voice (ElevenLabs…)?
Yes, in beta: choose “Spoken voice” before uploading your file. In our tests, 6 out of 10 ElevenLabs voices are caught, with 2% of real voices wrongly flagged (amateur readers included), although no ElevenLabs voice was used for training. Some synthetic voices, such as Microsoft neural voices, and the most recent cloned voices can still escape detection.
Is it free?
Yes. The analysis is free with no sign-up, up to 10 tracks per day. What you pay for is the Human Vibes certification, for artists and labels who want to prove that their music is human.
Does the report have legal value?
No. It is a decision-support tool that provides statistical clues. In a dispute, it must be complemented by a human expert assessment (listening, work sessions, source files).