Does a highly repetitive track mean it was AI-generated?

No. Repetition is a compositional and production norm: loop-based electronic music, library tracks and most pop choruses repeat heavily by design. Some generated output is unusually uniform, but so is a great deal of human loop-based production, so uniformity on its own is not evidence of origin.

What does 'near-identical passage' mean here?

It means two stretches of at least two seconds whose spectral fingerprints match almost exactly. In practice that is a copy-and-paste in a DAW, a looped sample, or a pattern rendered once and repeated. A live performance essentially never produces matches that tight.

How are section boundaries found?

The track is reduced to a fingerprint every 250 ms, and the tool compares the two seconds before each point with the two seconds after. Where the difference peaks, the spectral character has shifted — a verse-to-chorus edge, a breakdown, or an edit point where two pieces of audio were joined.

Why does the repeat cycle not match my tempo?

The cycle reported is whatever repeats most strongly, which is usually a bar group rather than a single bar — often four or eight bars. It is a structural period, not a beat-level tempo reading, and syncopated or half-time material can land on a multiple of what you expect.

Is my audio uploaded anywhere?

No. The file is decoded and measured by your own browser and released when you close the tab. Nothing is transmitted, stored or logged, so unreleased masters and NDA material are safe to check.

How much of the file is measured?

Up to five minutes, taken from the middle of the track so intros and fades do not dominate the figures. Shorter files are measured end to end.

Free tool

Song structure & repetition checker

Find a track's strongest repeating cycle, spot passages that are near-identical copies of each other, see how uniform the whole thing is, and get the points where the arrangement changes. Decoding and measurement happen in your browser — the file never leaves your machine.

Short answer

Can repetition tell you whether a song is AI-generated?

Not on its own. Heavy repetition is normal in loop-based electronic music, library tracks and pop choruses, and any generated track can be arranged to vary. What structure measurement does well is describe how a track was assembled — copied blocks, a single unbroken texture, or genuinely distinct sections — which is context for a detector score rather than a verdict.

Where detection misleads →

Drop an audio file here

MP3, WAV, FLAC, M4A, Ogg or Opus. Long files are measured on five minutes from the middle.

What each number means

Repeat cycle is the distance at which the track most strongly resembles itself. The audio is reduced to a coarse spectral fingerprint every 250 milliseconds, and every possible offset is scored by how well the fingerprints line up. The winner is usually a bar group — four or eight bars at the track's tempo — rather than a single bar, because that is the period at which arrangements genuinely recur.

Match strength is how closely the best-matching run holds together at that cycle. High values mean the repeat is very faithful; middling values mean the section returns but with different instrumentation or a variation on top.

Uniformity is the average similarity between fingerprints across the whole excerpt, so it describes the track as a texture rather than as a structure. A single unchanging groove reads high. An arrangement that moves through genuinely different worlds reads low.

Sections counts the regions between detected boundaries. A boundary is a point where the two seconds before it look substantially different from the two seconds after, and boundaries are kept at least four seconds apart so a busy fill does not register as a new section.

Why repetition is not an origin signal

The intuition is understandable: generated tracks are often described as loopy, structurally flat, or as four minutes of the same idea. Some are. Single-pass generation produces a whole stereo mix at once, and without an arranger making decisions there is no particular reason for a bridge to sound like a different room from the verse.

But repetition is also the backbone of most produced music. House and techno are built from loops on purpose. Library and production music is deliberately uniform so an editor can cut anywhere. Pop choruses repeat because that is what makes them choruses. Point this tool at a catalogue of human electronic music and you will find high uniformity and near-identical copied blocks everywhere, because DAWs make copy-and-paste the default way to build an arrangement.

So read structure as description, not as proof. Where it earns its place is in explaining a detector score and in directing attention: a track that is uniform, heavily copied and shows no section boundaries gives detection very little variety to work from, which is a reason to widen your uncertainty rather than narrow it. Combine it with file metadata, a spectrogram, a stereo reading and a stability check across excerpts.

Practical uses beyond provenance

Find the exact loop length of a track you want to edit or extend.
Locate section edges for a radio edit, a trailer cut or a sync submission.
Check whether a mix has any real variation or just repeats the same eight bars.
Spot an accidental duplicated section left in after an arrangement change.
Find the join in a track assembled from two different renders or takes.

Limits of this tool

Measurement is bounded to five minutes from the middle of a file, so a long DJ set is described by its middle rather than its whole. Fingerprints are coarse by design — 24 log-spaced bands with the overall level removed — which makes the comparison robust to volume changes but blind to fine detail such as a different vocal take over an identical backing. The reported cycle is one number for the whole excerpt, so a track that changes tempo or metre mid-way will report whichever period dominates. And boundary detection is a threshold on a curve: a gradual transition can be missed, and a dramatic fill can be counted.

For a fuller walk-through of cycle lengths, copied passages, uniformity and why repetition is a production technique rather than evidence of origin, read Song structure, repetition and AI music.

Questions

Does a highly repetitive track mean it was AI-generated?: No. Repetition is a compositional and production norm: loop-based electronic music, library tracks and most pop choruses repeat heavily by design. Some generated output is unusually uniform, but so is a great deal of human loop-based production, so uniformity on its own is not evidence of origin.
What does 'near-identical passage' mean here?: It means two stretches of at least two seconds whose spectral fingerprints match almost exactly. In practice that is a copy-and-paste in a DAW, a looped sample, or a pattern rendered once and repeated. A live performance essentially never produces matches that tight.
How are section boundaries found?: The track is reduced to a fingerprint every 250 ms, and the tool compares the two seconds before each point with the two seconds after. Where the difference peaks, the spectral character has shifted — a verse-to-chorus edge, a breakdown, or an edit point where two pieces of audio were joined.
Why does the repeat cycle not match my tempo?: The cycle reported is whatever repeats most strongly, which is usually a bar group rather than a single bar — often four or eight bars. It is a structural period, not a beat-level tempo reading, and syncopated or half-time material can land on a multiple of what you expect.
Is my audio uploaded anywhere?: No. The file is decoded and measured by your own browser and released when you close the tab. Nothing is transmitted, stored or logged, so unreleased masters and NDA material are safe to check.
How much of the file is measured?: Up to five minutes, taken from the middle of the track so intros and fades do not dominate the figures. Shorter files are measured end to end.