Is this audio AI-generated?
Drop an mp3, wav, m4a, ogg, or flac. We compute a 1024-bin STFT spectrogram in your browser and surface three forensic signals: spectral flatness uniformity, harmonic stability, and centroid drift. Nothing is uploaded.
MVP detector. Vocoder fingerprinting and phase coherence land next.
Spectrogram first
Real speech and music have rich, varied spectral structure with clear formants and transients. Vocoded audio often produces smoother, more uniform spectrograms — a visible tell once you know what to look for.
The signals are heuristic
No detector keeps up with the latest TTS / voice-clone models on raw accuracy. The point of forensic visualization is the same as for image: auditable, not magical. You see every signal that fires.
Source matters
C2PA-signed audio (Adobe, AI vendors, news organizations) reads cleanly when present. Until that's universal, layered spectral forensics is the bridge.