Voice clones & TTS
AI voice detector
Got a voice note from a family member asking for money, or a voicemail from your “boss”? Drop the recording here to check it for signs of text-to-speech and voice cloning — then read the advice below before you act.
Free · no sign-up · your file never leaves your device
Drop the voice message or recording or click to choose
MP3 · WAV · M4A · FLAC · OGG · Opus — the first 60 seconds are analysed on your device
How voice-clone scams work
A scammer finds a few seconds of someone’s voice — a social media video, a voicemail greeting, a podcast — and feeds it to a cloning tool. The tool generates new speech in that voice from typed text. The call or voice note is designed to rush you: an accident, an arrest, a kidnapping, a boss who needs an urgent transfer before a deadline. The voice is often slightly muffled or emotional, which hides imperfections, and the request comes with a reason not to call back.
The same technology impersonates executives in “CEO fraud”, politicians in robocalls before elections, and customers when calling banks that use voice verification.
What the detector checks in a voice recording
- The format. Mono WAV files at 16, 22.05 or 24 kHz are the default output of many speech engines. Messaging-app formats are recognised and handled separately.
- A bandwidth ceiling. Speech synthesised at a low internal rate and saved into a higher-rate file shows an abrupt cut-off in the spectrum at 8, 11 or 12 kHz. Telephone bandwidth (around 3.4–4 kHz) is recognised as a different, innocent cause.
- Perfect silence. Pauses that are exact digital zero, rather than quiet room noise, are common in synthetic speech and rare in real recordings.
- Noise floor. How loud the quietest moments are. Real phones pick up hiss, traffic, wind and breath.
- Pitch smoothness and evenness. Natural voices waver from one moment to the next; many synthetic voices are smoother and hold a more even loudness.
The result lists which of these fired and in which direction. The spectrogram above the player lets you see the cut-off and the silences for yourself: a real recording usually shows a hazy floor of noise in every gap, while synthetic speech often shows pure black between words.
Better than any detector: a callback
For a suspected scam call, detection is the second line of defence. The first is verification through a channel the scammer doesn’t control: call the person back on a saved number, message them on a different app, or ask a question only they could answer. Many families now agree a code word for emergencies. Banks and employers increasingly require a second confirmation for any transfer requested by phone — insist on it.
What ElevenLabs and Other TTS / voice-clone tool leave in a file
These are the traces GPTTrace checks for. “Some files only” means the trace is often missing — a re-save, screenshot or social-media upload removes metadata — so its absence proves nothing.
| Generator | Where | What to look for | How reliable |
|---|---|---|---|
| ElevenLabs | Audio | Studio-clean speech with very low noise floor and few breaths | Commonly seen |
| ElevenLabs | Container | MP3 at 44.1 kHz from the website; tags rarely name ElevenLabs | Commonly seen |