Suno detector

How Suno songs spread

Suno generates a full song — lyrics, vocals, instruments and mix — from a short prompt, and lets users extend, remix and cover tracks. Its quality rose quickly through version 4 and beyond, to the point where listeners frequently can’t tell. Songs made with it are uploaded to streaming services under invented artist names, used as background music in videos, entered in contests and sent around as jokes impersonating real artists.

Platforms have responded unevenly. Some label or demote fully AI-generated tracks, some require disclosure, and some remove tracks that impersonate real artists. Listeners are often left to work it out themselves.

What GPTTrace can check

Tags and download origin

Files downloaded straight from Suno can carry tags such as a comment, a source URL or a title that point back to the service. GPTTrace distinguishes tags that describe the software that made the file from free-text fields like song titles, so a song merely titled “Suno” isn’t counted as evidence. After a song passes through a distributor and a streaming service, these tags are usually gone.

Spectral traces

Generated audio is decoded from a compressed internal representation. In practice this can show up as a soft or abruptly ending top octave, smeared transients, and narrow tones in the upper band. GPTTrace measures each and shows the spectrogram; the top quarter of the image is where to look.

Context

For an unknown artist, look at the catalogue. A brand-new act with dozens of releases in a few months, generic cover art, no live dates, no interviews and no social presence is a common pattern for AI-generated catalogues.

Because the speech-specific checks don’t apply to music, results for songs depend on fewer signals than results for voice recordings. Treat a high score as meaningful and a low one as unproven.

Covers, remixes and voice imitation

Suno can extend existing audio and generate covers in a different style, and other tools convert one singer’s voice into another’s. That produces hybrids: a real backing track with a generated vocal, or a human vocal over a generated arrangement. Whole-track signal analysis averages these together, so a hybrid may score in the middle. If one element is suspicious — usually the vocal — and you can isolate it with a stem-separation tool, checking the isolated stem gives a clearer signal.

Voice imitation of famous artists is the case most likely to cause legal and reputational trouble. If a supposedly unreleased song by a well-known singer appears out of nowhere, check the artist’s official channels before sharing it; labels and artists usually respond quickly to viral fakes.