AI Music Detector.
Check your song.

The detector looks for spectral patterns characteristic of generators it knows. The final score selects one of six descriptions: human-made (below 20%), human-like patterns prevail (20–40%), minor AI patterns (40–50%), AI patterns (50–68%), strong AI patterns (68–85%), or rich in AI patterns (85–100%). These are interpretation bands, not measurements of how much AI was used. Each boundary starts the next band.

The AI pattern detection algorithm is based on the open-source project lofcz/ai-music-detector ↗.

Mastering, encoding, unusual instruments and new generators can affect the result. The track score uses the first five minutes. The timeline uses 60-second windows with a five-second step across the entire track. Its average is weighted by the duration of each mark, excluding silence and unavailable scores. The displayed score is (track score + timeline average) / 2. This summary is not a calibrated probability of the track’s origin. The model was trained on whole tracks; temporal localisation has not been independently validated.

Use this as a listening aid, not proof of a recording’s origin. Treat scores close to 50% with particular care.