Authenticity & AI
AI audio detector
Estimate whether a voice or audio clip was generated with artificial intelligence.
What you upload
MP3, M4A, WAV, OGG, OPUS, WebM and MP4
Limit
Up to 60 MB
Your file
Deleted as soon as it finishes
Takes
A few seconds to 2 minutes
- 1Upload
- 2Confirm
- 3Result
Your files are deleted as soon as the process finishes.
AI audio detector
Upload an audio clip or voice note and get an estimate of whether it sounds like a recorded human voice or one generated with artificial intelligence. It helps when a voice message makes you doubt and you need a first opinion. You can choose whether to check a voice or music and singing; the first 90 seconds are analyzed by a specialized detector. You get a verdict in three levels, a confidence rating and the signals that were observed.
What you get
- A verdict in three levels: probably a person, cannot be determined, or probably AI-generated.
- A confidence rating (low, medium or high).
- The signals observed in that specific clip, in plain language.
- The limits of the analysis next to the result, so it is not taken as proof.
How it works
- 1
Upload the audio
It accepts MP3, M4A, WAV, OGG, OPUS, WebM and MP4 (the audio track of a video), up to 60 MB. One clip is analyzed at a time, and you choose whether to check a voice or music and singing.
- 2
A specialized detector analyzes it
Only the first 90 seconds are analyzed: your browser trims the audio before uploading it and the result says so. If what matters comes later, trim the clip to that part.
- 3
Read the result
You see the verdict, the confidence and the signals. If the specialized detector cannot analyze the file, no verdict is given: the result is "cannot be determined" and only general observations are shown.
- 4
Verify another way
If the clip asks for money, passwords or an urgent decision, confirm with the person through a different channel before acting.
When is it useful?
- People who get a voice message from a relative or boss asking for money or data and have doubts.
- Journalists and fact-checkers reviewing a clip attributed to a public figure.
- Support or fraud teams who receive recordings and want a first opinion.
- Anyone deciding whether a clip is worth investigating further before sharing it.
Keep in mind
- It is an automatic estimate and can be wrong in both directions: it may flag a real voice as AI or let a synthetic one through.
- Only the first 90 seconds of the clip are analyzed.
- Background noise, compression and phone recordings make the analysis harder.
- Recent synthetic voices can sound natural, even to an attentive ear.
- It is guidance, not proof: do not use it as the only basis to accuse or penalize someone.
- A result of "probably a person" does not prove who is speaking or that what is said is true.
What happens to your file?
The audio is uploaded to a private store only to be analyzed and is deleted as soon as the analysis ends. Its content is not kept in logs. Privacy policy.
Frequently asked questions
How can I tell if a voice or audio clip is AI-generated?
There is no foolproof method. Listening for breathing, rhythm and cuts helps, but the most useful step is confirming who sent it and verifying the request through another channel. This tool runs the first 90 seconds of the clip through a specialized detector and summarizes what it observed into an estimate.
How reliable is an AI voice detector?
It is guidance, not proof. It can fail in both directions and does worse with noisy or compressed audio. That is why it always shows the confidence and the limits of the analysis.
Is it better than just listening carefully?
It gives a second opinion and writes down the signals it observed. But there is one thing no analysis replaces: calling the person and confirming through another channel that the message is theirs.
Is my audio uploaded anywhere?
Yes: to be analyzed, the file goes to a private store and is deleted when processing ends. Do not upload it if it contains information you are not entitled to share.
Can it prove a voice is fake?
No. It is an automatic estimate. For anything with consequences, use it as a starting point and rely on the file's origin and a forensic expert.
Which formats does it accept?
MP3, M4A, WAV, OGG, OPUS, WebM and MP4 (the video's audio track is used), up to 60 MB, one clip at a time. The first 90 seconds are analyzed.