Upload audio — get a verdict in 10 seconds. Detects ElevenLabs, Murf, PlayHT, OpenAI, and other major AI voice tools. Free, no signup.
AI voice cloning is now indistinguishable from human speech to most listeners. ElevenLabs can clone a voice from 3 minutes of audio. OpenAI's TTS reads any text in a chosen voice. Murf and PlayHT are used to narrate ads, podcasts, and customer-service messages.
The applications are useful — accessible audiobooks, e-learning, automated phone systems. The misuse is concerning — voice phishing scams, deepfakes for political manipulation, fake celebrity endorsements, fabricated audio "evidence."
TextSight's AI Voice Detector analyzes spectral features, breathing patterns, micro-pause structures, and waveform characteristics that AI voice models leave behind. Upload an audio file, get a verdict in 10 seconds.
Yes, 3 audio checks per day free with no signup. Sign up free for 20/day.
MP3, WAV, M4A, OGG, FLAC. Max 10 MB or roughly 5 minutes per file.
ElevenLabs (all models), OpenAI TTS, Murf, PlayHT, Amazon Polly, Microsoft Azure TTS, Google Cloud TTS, Resemble, Descript Overdub, Adobe Speech-to-Speech, and most consumer AI voice apps.
~89-93% on clean studio AI audio, ~80-85% on AI audio with added background noise or compression, ~95% on real human speech. Phone recordings are hardest because they share some compression artifacts with AI audio.
No. Files are processed and discarded immediately. We don't store them, train on them, or share them.
Yes. Spectral analysis is language-independent. Tested across English, Hindi, Spanish, French, Mandarin, Japanese, Arabic.
Run it through TextSight Detector — free, 3 scans/day.