Detect cloned voices & AI-generated audio
Find out if a voice message, phone call or recording is a real person or an AI clone. ID/Ai listens to the sound itself, so it works on phone calls, voice notes and clips that have been shared around.
- Results in seconds
- Works on phone-quality audio
- Shareable certificate
- 1 free check, no sign-up
Every recording is checked for the fingerprints of AI voices
Voice cloning now needs only a short sample of someone speaking, and the results fool most listeners. Our models hear what people can’t, and show you what each one found.
- Voice cloningCopies of a real person’s voice made from a short sample, the tool behind most voice scams.
- Text-to-speechFully synthetic voices from AI speech services, now used for fake ads, robocalls and news.
- Open-source voice modelsFree tools anyone can run on a laptop, often with no safeguards or watermarks.
- AI processingNoise cancelling and clean-up applied to a real voice, scored separately so it isn’t mistaken for a fake.
- Splicing & editsWords cut, rearranged or inserted to change what someone appears to have said.
- Metadata & ledgerThe file’s hidden history and whether a verified original is on record.
Synthetic speech throughout, with no sign of normal call processing.
- AI-generated speech96%
- Voice cloning match95%
- Text-to-speech match4%
- AI-processed (noise removal etc.)5%
- Splicing & editsNone found
- LedgerNo verified original
Cleaned-up audio isn’t fake audio
Zoom noise cancelling, phone compression and podcast clean-up tools all use AI on real voices. We score AI-generated and AI-processed separately, so a real person on a noisy call isn’t wrongly called a fake.
A clean recording of a real person.
A real caller with noise cancelling switched on. Many detectors wrongly call this fake.
Speech created by a voice-cloning tool.
Don’t trust a voice just because it sounds right
Scammers can clone a voice from a short clip posted online, then use it to ask for money, passwords or access.
Check that panicked “it’s me, I need money” call or voice note really is your family member.
Stop cloned “boss” voices asking staff to urgently pay an invoice or change bank details.
Protect voice-based identity checks and contact centres from synthetic callers.
Catch fake endorsements using the cloned voices of well-known Australians.
Verify “leaked” recordings and robocalls before they sway an election or a news story.
Get evidence that a recording of you saying something you never said was faked.
Trained to spot the tools people actually use
New generators are added as they appear.Trademarks belong to their respective owners. These companies are not affiliated with ID/Ai.
Multiple layers of proof, not one guess
Social platforms strip metadata and most AI audio carries no watermark. So we don’t rely on any single signal.
AI detection
Models analyse the sound wave itself, so they still work on phone calls, voice notes and clips that have been compressed or re-shared.
Metadata & forensics
We read the file’s hidden history: recording device, editing software, compression and content credentials, where they exist.
Blockchain ledger
We check whether a verified original is on record, and every result gets a timestamped certificate you can share.
Drag, drop, done
Perfect for checking a voicemail or recording before you act on it. Get a report you can screenshot, download or send to your bank.
- Results in seconds
- Works on phone-quality audio
- Shareable certificate
Plug it into your platform
Screen calls and voice messages at scale in contact centres, voice ID checks and newsrooms with the ID/Ai API.
curl -X POST 'https://api.idai.ai/v1/check' \
-H 'Authorization: Bearer {API_KEY}' \
-F 'media=@voicemail.m4a' \
-F 'checks=ai_speech,ai_processed'
How can you tell if a voice is AI?
AI voices leave tiny patterns in the sound wave, in breathing, pitch and background noise, that the human ear can’t pick up. Our models are trained to recognise them, then we combine the result with metadata and our verification ledger.
Does it work on phone calls and WhatsApp voice notes?
Yes. It’s built for real-world audio, including phone-quality calls and compressed voice messages. Very short or very noisy clips can give a less certain result.
My call app uses noise cancelling. Will that be flagged as fake?
It shouldn’t be. We score AI-generated speech and AI-processed audio separately, so a real voice that’s been cleaned up is reported as real.
How long does the recording need to be?
A few seconds of clear speech is usually enough for a result. Silence is left out, so it’s the speech that counts. The detector on this page takes files of up to 40 MB.
I think I’ve been targeted by a voice-clone scam. What should I do?
Don’t send money or share details. Hang up and call the person back on a number you know. Download your ID/Ai report, contact your bank straight away, and report it to Scamwatch and ReportCyber.