How to verify if an audio is generated by AI.
A short, plain-English guide to running a verdict on suspicious audio.
What is aivoicedetector.com?
A web service that takes an audio clip and returns a probability that it was generated by a synthesis model. The detector covers more than 24 generators in production, including ElevenLabs, Resemble, PlayHT, and OpenAI's TTS. It runs in roughly 480 milliseconds. Free for a single verdict.
Why use it?
Three reasons. First, the human ear is not reliable for current-generation synthetic voices. Second, decisions made on bad audio carry real costs (publishing a story, filing a brief, sending a wire). Third, a citable verdict travels: you can quote it to an editor, a court, or a board.
How does it work?
The detector analyzes the acoustic fingerprint of the audio. Real human voices and synthesized voices leave different traces in spectral, temporal, and prosodic dimensions. We trained on millions of examples from both sides and we update the model monthly as new synthesis systems ship. The full methodology is in our methodology post.
Step by step
- Save the audio file (do not screen-record).
- Open /is-this-ai and drop the file.
- Read the verdict, confidence number, and any model attribution.
- If it matters, save the verdict to your dossier for a citable URL.
Where we are wrong
Heavily compressed audio (phone calls, low-bitrate WhatsApp voice notes) lowers our confidence. So do very short clips (under half a second of clear speech). For those, we report a lower confidence number rather than guess. We tell you when we cannot tell.
We return a probability with a method behind it. Not a verdict on the speaker. A verdict on the audio.