Free check, no account

AI Voice Detector. Detect AI-generated voices.

Upload a voice note, call recording or voice clip. You get a verdict, an AI probability score, and a breakdown showing which parts of the clip sound synthetic.

Open detector See pricing
Example result
Verdict
Likely AI.
Flagged 0:12
AI score
72%
Segments
2 of 6 flagged
Try the detector on your audio →
As featured in
ForbesNPRUnivisionWiredCybernews
Who uses it

Who uses AI Voice Detector.

From newsrooms checking a clip before publication to banks reviewing a flagged call, and families checking a voicemail.

Fig 1.1
Newsrooms

Journalists

Check suspicious audio before publication, save the result, and share the link with your editor. For newsrooms →

Fig 1.2
Federal . legal

Investigators

Keep a dated record of every check in your history, with share links that do not expire. Use the result alongside your other evidence. For legal & investigations →

Fig 1.3
Compliance

Banks & call centers

Check a suspicious call recording before the payment goes out, or run checks through the API. In the US, the FCC ruled in 2024 that AI-generated voices in robocalls count as "artificial" under the TCPA, the same consent rules that govern prerecorded calls. Fraud detection use case →

Fig 1.4
Individuals

Families and individuals

A voicemail from a relative asking for money, or a call from "your bank". Check it before you call back or pay. No account needed.

Features

What you get.

One model behind the web detector and the API. Every check returns a verdict and an AI probability score.

Fig 0.1

Upload and check

Upload an MP3, WAV, M4A, OGG, FLAC or AAC file and get a verdict and an AI probability score.

Fig 0.2

Segment scores

Longer clips are split into segments and each one is scored, so you can see which part of a recording sounds synthetic.

Fig 0.3

Share links

Save a result to your history and share it with a link that does not expire.

Fig 0.4

History and CSV export

Every check you run in your account is kept in your history. Export it to CSV on Starter and above.

Fig 0.5

REST API

Upload a clip and get the verdict, the score and the segment scores back as JSON. Same model as the web detector. Starter and above.

Fig 0.6

Chrome extension

A new extension for checking audio from any tab is in development. Until it ships, use the web detector.

At a glance

As featured in Forbes, NPR, Univision, Wired and Cybernews.

The detector, the formats it reads and what every check returns.

Press
Five
Outlets that have featured AI Voice Detector.
Formats
MP3 . WAV . M4A
Plus OGG, FLAC and AAC.
Free tier
50 a month
Free checks, no card required.
Every verdict
Four-step verdict
AI Generated, Likely AI, Likely Human or Human, plus an AI probability score.
Three steps

How it works.

No account needed to use the free AI voice detector. Sign in when you want to save results to your history or share them.

01 . Upload

Upload a clip.

Drag in an audio file with at least 3 seconds of clear speech.

MP3 . WAV . M4A . OGG . FLAC . AAC
02 . Analyze

We score it.

The model scores the clip segment by segment and returns an AI probability score.

AI score . segment scores
03 . Read

Read the verdict.

See the verdict on a four-step scale and which segments sound synthetic. Save it to your history or share the link.

Verdict . share link
Landscape

The AI voice tools in use today.

The services and models most likely behind a synthetic voice you hear now, from commercial cloning platforms to open-source models anyone can run. The detector tells you whether a voice is AI-generated. It does not identify which of these tools made it.

ElevenLabsv3 . cloning CartesiaSonic . real-time
xAIGrok voice . TTS
MiniMaxSpeech . cloning
Fish AudioOpenAudio S1
Resemble AIVoice cloning OpenAIgpt-4o-mini-tts GoogleGemini TTS . Chirp 3 HD MicrosoftMAI-Voice-1 . Azure Amazon PollyGenerative TTS Murf AIStudio TTS WellSaid LabsStudio TTS SpeechifyRead-aloud TTS DescriptEditing . cloning
Qwen3-TTSAlibaba . voice cloning
ChatterboxOpen source . cloning
CosyVoice 2Alibaba . open source
DiaNari Labs . dialogue
KokoroOpen source . lightweight
OrpheusOpen source . Llama-based
SunoMusic . vocals
UdioMusic . vocals
Guides by generator →Deepfake voice detection →
Reading the result

What each verdict means.

Provenance standards such as C2PA can prove a genuine recording's origin, but only when the file still carries its manifest, and most forwarded audio does not. The human ear is unreliable on current-generation synthesis. The FBI agrees: AI-generated content "is often difficult to identify". The detector gives you a probability score and a graded verdict instead of a guess. Here is how to read it.

VerdictAI scoreWhat to do
AI Generated86 to 100%Treat the voice as synthetic. Do not act on the request; confirm with the person through a number you already know.
Likely AI50 to 85%Probably synthetic. Verify through another channel before you pay, publish or reply.
Likely Human16 to 49%Probably real, but not proof. Weigh it with the context: who sent it, and what it asks for.
Human0 to 15%No sign of synthesis in the audio. Still check the request if money or reputation is at stake.
Short or low-quality clipWarning shownResults are less reliable. Get a longer or cleaner recording and check it again.
For developers

Developer API.

The same model behind a REST API. Upload a clip, then get a verdict, an AI probability score and per-segment scores back as JSON. Included on Starter and above.

Explore the API →
POST /api/v1/detect/voice 
curl -X POST .../api/v1/detect/voice \
 -d '{"upload_id": "..."}'

→ 200 OK
{
 "verdict": "likely_ai",
 "ai_score_percent": 72,
 "segment_count": 6
}

We don't think AI voices are going to slow down. We think anyone holding a recording should be able to ask "is this real?" and get a clear answer.

Every check returns a verdict on a four-step scale and an AI probability score. Longer clips are scored segment by segment, so you can see where a recording sounds synthetic. Results can be saved to your history and shared with a link. And the team behind it answers email.

The team
AI Voice Detector
How it works
From the blog

Guides.

All articles →
Scams

The grandparent scam, now with AI voice cloning

How a cloned voice poses as a relative in trouble, the red flags, and how to stop it.

Jul 202610 min
 Fraud

CEO fraud: how voice cloning defeats finance teams

The anatomy of the attack, the red flags, and the controls that stop it.

Jul 202611 min
 Guide

How to detect AI voices. Voice cloning and audio deepfakes.

A fast checklist, the tells to listen for, and how to confirm with a detector.

Updated Jul 202614 min
FAQ

Frequently asked questions.

Can't find your answer? Email support@aivoicedetector.com.

How accurate is the detector?
Every check returns an AI probability score from 0 to 100% and a verdict on a four-step scale: 86% and above is labelled AI Generated, 50 to 85% Likely AI, 16 to 49% Likely Human, and below 16% Human. A Likely AI or Likely Human verdict is borderline, so treat it as one piece of evidence. Short, compressed or noisy clips, such as phone calls and low-bitrate voice notes, give less reliable results.
How long does a clip need to be?
At least 3 seconds of clear speech. The app warns you when a clip is shorter than that or recorded at low quality. Free accounts can check clips up to 2 minutes and paid plans up to 10 minutes, with files up to 50 MB.
What happens to my audio?
The free check needs no account: the audio is analysed in memory and discarded. When you run a check in your account, the recording and its result are kept in your history so you can open and share them later. See the privacy policy for details.
Can I use a result in journalism or legal work?
Yes, as one piece of evidence. Save the result to your history and share it with a link that does not expire, so an editor or a court can see what was checked and what it returned. Pair it with sourcing and, where you can, the original file.
Is there a free tier?
Yes. 50 detections a month, no card required, with clips up to 2 minutes. Every check returns a verdict and an AI probability score. See pricing for paid plans.
Do you have an API?
Yes, on Starter and above. Upload a clip, then request a detection, and get the verdict, the AI score and per-segment scores back as JSON. Same model as the web detector. If you need more volume, email the team.

Check a recording now.

Free, no account needed.

Open detector How it works