A step-by-step look at the acoustic analysis behind AISongScan and why it works.
You upload an audio file in MP3, WAV, M4A, or FLAC format. Your file is transmitted over an encrypted connection.
The audio is converted into a mathematical representation of its acoustic characteristics -- timing patterns, spectral data, pitch variation, dynamic range, and transient behavior.
The fingerprint is analyzed against known signatures of AI-generated audio. Our detection system evaluates timing consistency, harmonic structure, synthetic smoothness, and other markers that separate machine-generated audio from human performance.
Where possible, the system identifies which AI generation tool -- Suno, Udio, Sonauto, Mureka, Riffusion -- most closely matches your track's profile.
You receive an AI probability score between 0 and 100. A score above 50 indicates the track likely contains AI-generated audio. The score is a probabilistic signal, not a legal verdict.
AI-generated music leaves distinct acoustic markers that differ from human performance. Here is what our detection system analyzes:
Human performers exhibit natural micro-timing variation. AI generation tends to be unnaturally precise.
Organic recordings contain dynamic variation that AI tools often compress or regularize.
AI generation can introduce spectral patterns in specific frequency bands that differ from acoustic recordings.
Phase relationships between audio channels in AI tracks differ from those produced by live instruments and rooms.
Attack and decay patterns in AI-generated audio follow mathematical curves rather than physical acoustic behavior.
Overtone relationships in AI music are constructed mathematically and lack the complex variation of acoustic instruments.