Detect a specific AI voice.
Most detectors give a yes or no. We attribute the source across 24+ systems. Pick the generator you suspect, or drop any clip into the detector and let it name the model for you.
Start with the source you suspect
Coverage keeps pace: we add new systems as they ship and stamp the methodology version on every verdict. Not sure which made a clip? Drop it into the detector and it names the model, or returns an honest "unknown synthesis".
Why naming the generator matters
Most detectors stop at a yes or no. We name the system where we recognize it, because the source changes what you do next. A clip traced to a consumer voice-cloning tool points to impersonation; one from a cloud text-to-speech engine points to automation or a bot. For a fraud analyst, a journalist, or an investigator, the named model is a lead, not just a label, and it travels into a report or a case file in a way a bare percentage does not.
Attribution also keeps us honest. The detector reads the acoustic fingerprint a generator leaves behind, so it works on both cloned and stock voices, and when the signature is not distinctive enough to attribute, it returns an honest "unknown synthesis" rather than guessing a brand.
What the detector covers
The pages here span the systems people actually encounter: expressive voice-cloning tools like ElevenLabs, Resemble, PlayHT, and Descript; studio and read-aloud text-to-speech like Murf, WellSaid, and Speechify; the big cloud engines from Amazon, Microsoft, and Google; open-source models such as Coqui, Bark, and Tortoise; the VALL-E research line; and Suno for AI music and vocals. Each guide explains how that generator tends to sound, the tells worth listening for, and how the detector separates it from a real human voice. If your hunch is wrong it does not matter: the detector reads the audio, not your guess, and names whatever it actually recognizes. New systems are added the month they ship, and the methodology version is stamped on every verdict, so a result stays reproducible even as the models behind these voices keep changing. For the underlying method, see how the detector reads a clip; to check something now, open the detector and drop the file in.
The audio, not the speaker
The detector reads the acoustic signature a generator leaves behind, so it works on cloned and stock voices alike, and returns the recognized model or an honest "unknown synthesis" when it cannot attribute the source. Learn more about deepfake voice detection.
Drop the clip
Upload a file or paste a URL. MP3, WAV, M4A, WebM, or the audio track of a video. About half a second of clear speech is enough.
The model scores it
The same model behind the public detector reads the acoustic signature and weighs the artifacts, then attributes the source, e.g. the model, when it recognizes it.
Get a citable verdict
You get a probability, a confidence level, the named model, and a permanent citation URL you can quote, file, or subpoena.