"AI audio detector" covers two different jobs. One checks speech: was this voice cloned or synthesised? The other checks music: was this song, or part of it, made by a music generator such as Suno or Udio? The tools are trained on different audio and are not interchangeable. This page is about music.
Music detection and voice detection are different tools
A voice detector learns what synthetic speech sounds like. A music detector learns the traces that music generators leave across a full mix: codec residue, smeared transients and high-frequency patterns that recorded instruments rarely produce. A voice model run on a song, or a music model run on a podcast, is outside what it was trained for, so its score means little.
ArtifactNet, the detector behind Intrect's checker, is a music model. Use it for songs, stems and mixes, not for phone calls or voice notes.
What a music detector returns
ArtifactNet gives an automated estimate, from 0 to 1, that the audio was machine generated, and a verdict when that estimate crosses a decision threshold. It also scores every 4-second segment, so a human track with one generated section does not read the same as a wholly synthetic one.
It analyzes patterns in the audio itself. It does not ask the generator for a signal and does not look for a mark planted in the file.
How to read the result
- Treat the score as evidence for a human decision, not a ruling about any person.
- Look at the segments: a short run of high scores points to a section worth listening to again.
- Compare like with like: the same song as WAV and as a heavily compressed MP3 can score differently.
- Expect errors in both directions; false positives and false negatives are possible.
Check a song now
The free checker at demo.intrect.io takes one WAV, MP3 or FLAC file at a time with no signup, and returns a label and an AI probability. For many tracks, exports and an API, see ArtifactNet.
Check one track, free
No account, no card. Upload an MP3, WAV or FLAC and read the verdict, the probability and the segment timeline. The first analysis of the day shows the full panel.