Upload a song and find out whether it carries the fingerprint of Suno or Udio, the two most used SI (AI) music generators. You get a probability and the clues behind it in a few seconds. Free, no credit card, and the audio is deleted right after the analysis.
Uploading and analysing... decoding up to 5 minutes of audio and looking for the Suno or Udio fingerprint (a few seconds)
Songs made with SI (AI) generators now reach streaming playlists, video soundtracks and music competitions, often without a label. This SI music detector, part of the SI detector suite, answers a narrow question well rather than a broad one badly: does this file carry the fingerprint of Suno or Udio? It is also a simple Suno detector and Udio checker for anyone asking "is this song AI?".
Generators like Suno and Udio do not record instruments. They compose a song as a compact internal representation, then a decoder turns it into sound. That last step is where the fingerprint comes from.
The two signals are combined into one probability. The fingerprint does most of the work, because tags are easy to delete or edit.
When the fingerprint is clear, the verdict is Likely SI-generated music. We chose a cautious threshold so that real recordings are not flagged: we would rather miss a generated song than accuse a musician. We do not publish an accuracy figure for the beta, because it would depend entirely on the songs it was measured on.
This is a beta with a narrow scope. Here is what it will not catch.
That is why a negative result reads No SI fingerprint found, never "human-made". No Suno or Udio fingerprint found does not prove a human made it: our beta detector recognises Suno and Udio, not every music generator.
Above roughly 70% we show Likely SI-generated music, below 30% No SI fingerprint found, and in between Uncertain, which usually means a partial fingerprint blurred by compression or editing. The clues list tells you why: fingerprint found or not, generator named in the tags, short excerpt, low bitrate copy. The metadata panel shows the duration, codec, bitrate, sample rate and channels of the file.
Use the score as one clue among others. Before you call a song generated, look at the context too: an artist with no history who releases dozens of tracks a week, credits that name no musician, lyrics that feel generic, or cover art that looks generated. And before you call a song human, remember that the detector only knows two generators.
Your audio file is written outside the web server under a random name, analysed on our own server, then deleted immediately, whether the analysis succeeds or fails. Nobody listens to it and it is never shown on a result page, even a shared one: a share link shows only the score, the clues and the technical details of the file. We log the score and the clues, never the audio, never the file name and never your IP address in clear text. Details on the privacy page.
Upload the song and we decode up to its first five minutes, measure its frequency spectrum and look for a regular pattern of tiny peaks that the audio decoders of Suno and Udio leave in everything they generate. We also read the file tags, where Suno and Udio often write their name. You get a probability, the clues behind it and the technical details of the file.
Yes, those are the two generators it is built for. Suno songs, including recent versions, carry a very clear fingerprint. Udio songs are recognised too, but less often, especially after the file has been re-compressed. Songs from other generators are usually not recognised.
Not necessarily. It means we did not find the Suno or Udio fingerprint. A song made with another music generator, a cloned voice over real instruments, or a heavily re-encoded copy can all give this result. Treat it as "no evidence found", never as proof that a person wrote and performed the song.
No. The detector looks at the fingerprint of whole generated songs. A real recording with a cloned or synthetic voice on top, or a human singer over a generated backing track, is not specifically detected.
The fingerprint is averaged over the whole excerpt, so a longer piece gives a cleaner signal. Below 30 seconds the result is not reliable enough, so we refuse the file and nothing is charged. A full song is best.
No. Those pages stream audio through their own players and we cannot read a song from a web page. Download the file you want to check (for example the MP3 from Suno, Udio or your distributor) and upload it, or paste a direct link that ends in .mp3, .wav, .m4a, .ogg or .flac.
MP3, WAV, M4A, AAC, OGG and FLAC files up to 25 MB, with at least 30 seconds of audio. A music check uses 1 credit, like an image. Your first 3 checks need no account, a free account gives you 10 credits a month, Pro 250 and Max 1,500. If the analysis fails, the credit is given back.
No. The audio file is written to a private folder with a random name, analysed on our own server, then deleted right away, whether the analysis succeeds or fails. We keep the score and the clues, never the audio, and a shared result link never plays or shows the song.
People guess barely better than chance, and detectors trained on photos stumble on drawings and paintings. Here is what the research says, which clues really help, and how our own detector did on eleven artworks.
Read the articleC2PA Content Credentials are signed provenance data that record which tool made a file and how it was edited. Here is how they work, how to check them, and what happened when we removed them from ten ChatGPT photos.
Read the articleCloned voices, generated profile photos and SI-written messages give old scams new disguises. Here is what the FBI, the FTC and Europol document, what our detector could and could not catch, and the checks that actually protect you.
Read the article
