Does an unusually wide stereo image mean a track is AI-generated?

No. Stereo width is a mixing and mastering decision applied after the music exists, and any track can be widened or narrowed in seconds. Some generators do produce a characteristically even stereo spread, but so do plugin-widened human mixes, so width on its own is not evidence of origin.

What does inter-channel correlation actually measure?

It compares the two channels sample by sample and returns a value between -1 and +1. Around +1 means the channels are nearly identical, so the image is centred and mono-safe. Near 0 means they carry largely independent content. Negative values mean parts of the signal cancel when the channels are summed.

Why does mono compatibility matter?

A great deal of real playback sums to mono or something close to it: single-speaker phones, many Bluetooth speakers, club systems, and some broadcast chains. If your master loses several decibels when folded, listeners on those systems hear a thinner, quieter mix than you approved.

Should bass be in stereo?

Usually not much. Wide low frequencies are the most common cause of mono cancellation and of unpredictable behaviour on large systems, which is why most engineers keep everything below roughly 150 Hz near the centre. The per-band table here shows how much side energy your low end carries.

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

Stereo field & phase checker

Measure how wide a track really is: side energy, inter-channel correlation, width per frequency band, channel balance, and how much level survives a fold to mono. Decoding and measurement happen in your browser — the file never leaves your machine.

Short answer

Can stereo width tell you whether a song is AI-generated?

No. Width and phase are set by mixing and mastering choices made after the music exists, and a plugin can change either in seconds. What these numbers are good for is explaining a detector score: an unusually narrow or heavily widened master changes the spectral detail detection reads, which is a false-positive risk rather than evidence.

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

Side energy comes from splitting the two channels into a mid signal, which is what both channels share, and a side signal, which is what they differ by. The percentage is the share of total energy carried by that difference. Most commercial masters sit somewhere between 2% and 15%. Very low values mean an almost centred image; very high values usually mean a widener, a Haas delay, or a mid/side balance mistake.

Correlation compares the channels sample by sample. A value near +1 means they are nearly the same signal, so the image is narrow and completely mono-safe. Values near 0 mean the two sides carry largely independent content, which is normal for wide ambient and electronic material. Negative values mean part of the signal cancels when summed — sometimes deliberate, often an inverted channel somewhere in the chain.

Side versus mid expresses the same relationship in decibels, which is the way most mastering tools show it. Mono fold is the change in level when the channels are summed: zero means nothing cancels, and several negative decibels means a noticeable amount of the mix disappears on single-speaker playback.

Why width is not an origin signal

It is tempting to treat an oddly even, glossy stereo image as an AI tell, and some generated output genuinely does have one — a wide, homogeneous spread with little movement between sections, because the whole track was rendered in one pass rather than assembled from separately panned parts. But that description also fits a great deal of plugin-widened human production, and it disappears the moment somebody re-masters the file. Width is one of the cheapest properties of an audio file to change, which makes it one of the worst to build a provenance claim on.

Where these measurements earn their place is as context. Detection reads spectral and transient detail, and both a collapsed image and an aggressively widened one alter that detail. If a track reads high and also shows an extreme stereo picture, the honest conclusion is that your uncertainty should widen, not narrow. Combine this with file metadata, a spectrogram, a loudness reading and, when the stakes are real, a direct request for stems or session files.

Practical mixing checks this answers

Is my low end wide enough to cancel on a club system or a phone speaker?
Has a channel been inverted somewhere between the mix and the master?
Did the stereo widener on the master bus go further than I intended?
Is the whole mix leaning left or right because of a panning slip?
Does the image actually open up in the chorus, or does it only feel like it does?

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. Band splitting uses second-order filters with gradual slopes, so the per-band figures are indicative rather than surgical. Correlation is computed over the whole excerpt, which means a track that is narrow for two minutes and wide for two minutes reports something in between — read the time trace alongside it. And the browser's own decoder reconstructs lossy files, so a heavily compressed MP3 can read marginally differently from the master it came from.

For a fuller walk-through of mid/side energy, correlation sign, mono cancellation and why width makes such poor provenance evidence, read Stereo width, phase and AI music.

Questions

Does an unusually wide stereo image mean a track is AI-generated?: No. Stereo width is a mixing and mastering decision applied after the music exists, and any track can be widened or narrowed in seconds. Some generators do produce a characteristically even stereo spread, but so do plugin-widened human mixes, so width on its own is not evidence of origin.
What does inter-channel correlation actually measure?: It compares the two channels sample by sample and returns a value between -1 and +1. Around +1 means the channels are nearly identical, so the image is centred and mono-safe. Near 0 means they carry largely independent content. Negative values mean parts of the signal cancel when the channels are summed.
Why does mono compatibility matter?: A great deal of real playback sums to mono or something close to it: single-speaker phones, many Bluetooth speakers, club systems, and some broadcast chains. If your master loses several decibels when folded, listeners on those systems hear a thinner, quieter mix than you approved.
Should bass be in stereo?: Usually not much. Wide low frequencies are the most common cause of mono cancellation and of unpredictable behaviour on large systems, which is why most engineers keep everything below roughly 150 Hz near the centre. The per-band table here shows how much side energy your low end carries.
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.