Back
Audio Deepfake Detector

Audio Deepfake Detector

Upload any audio file — MP3, WAV, M4A, OGG, FLAC, and more. Up to 12 acoustic and deep DSP checks measure pitch, spectral dynamics, and voice periodicity to estimate if the voice is AI-generated. No external APIs, no data sent to third parties.

Detection Model

Drop an audio file here

MP3, WAV, M4A, OGG, FLAC · Max 50 MB

How detection works

1. Prosody Analysis

Silence ratio, pause rhythm, and pitch standard deviation are measured across 20–100ms frames. TTS systems produce unnaturally dense speech with flat intonation contours.

2. Spectral Fingerprinting

ZCR variance, spectral flux, energy dynamics, MFCC trajectory smoothness, cepstral peak prominence (CPP), HNR, micro-timing jitter, and centroid drift. Neural vocoders leave characteristic signatures.

3. Accent-Aware Scoring

Up to 24 indicators are weighted by signal strength. 7 prosodic checks are marked accent-sensitive — their thresholds are relaxed for non-native English speakers. If only prosodic flags fire while physiological checks (jitter, shimmer, HNR) are clean, the score is automatically reduced.