ElevenLabs voice detector

Where ElevenLabs voices show up

ElevenLabs became the default voice engine for a large part of the internet: faceless YouTube channels, TikTok narration, audiobooks, dubbing, podcast intros, video game characters and customer-service bots. Its voice library includes thousands of stock voices, and its cloning lets anyone recreate a voice from samples. Most of this is legitimate. The same tools also produced some of the best-known audio deepfakes — fake political robocalls, impersonated celebrities and cloned relatives in phone scams.

Recognising the stock voices is a skill some people develop: a handful of popular narrator voices are heard across thousands of channels. But clones of specific people are another matter, and that is where automatic checks help.

What GPTTrace can see in ElevenLabs audio

Tags

Files downloaded straight from ElevenLabs usually carry little metadata. If an encoder, comment or URL field names ElevenLabs — or a C2PA manifest is present — GPTTrace reports it. Most real-world copies have been re-encoded by a video editor or a messaging app, so expect tags to be missing.

Studio-clean signal

ElevenLabs speech is typically very clean: no room reverberation, no background noise and pauses with a very low noise floor. Real recordings, even from good microphones, almost always carry some room tone. GPTTrace measures the floor and any stretches of exact digital silence.

Prosody statistics

GPTTrace tracks pitch through voiced speech and reports how smoothly it moves and how even the loudness is. ElevenLabs’ expressive models vary both more than older TTS, so these checks are weaker on its newest voices — the result will tell you which signals fired and how strongly.

Recommended checks

1. Run the file here first — it’s private and takes seconds.
2. If the result is inconclusive and the audio matters, use ElevenLabs’ own classifier, which can recognise output from its models.
3. Where the audio claims to be from a real person, verify with that person directly. No detector can rule out a clone from another service.

Listening tips for ElevenLabs voices

Listen to the breaths. Real speakers breathe at phrase boundaries with audible air and room sound; generated breaths are often too regular or entirely absent. Listen to long numbers, names and unusual words, where synthetic prosody can stumble, and to emotional peaks, where the voice may become strangely smooth. If you have a genuine recording of the person for comparison, play both: clones capture timbre well but often miss a speaker’s particular rhythm, pauses and verbal habits.