Audio Channels changes the relationship between the left and right channels of stereo audio: you can swap them, merge them into mono, export the left and right channels separately, or subtract one channel from the other to try to thin out a vocal sitting in the centre. It lives on the “Channels” tab of the audio/video workbench, where selecting audio shows the four operations straight away.

“Split stereo” produces two output files at once, and “Karaoke vocal cut” is not AI source separation: it exploits the fact that in many stereo mixes a centred voice appears almost equally in both channels and subtracts them, so it only works on some songs. All processing is done by single-threaded ffmpeg.wasm inside the browser.

1. 01Add audiodrag audio into the audio/video workbench, select it on the left and switch to the “Channels” tab.
2. 02Choose an operationclick “Swap left/right”, “Merge to mono”, “Split stereo” or “Karaoke vocal cut”; the interface warns when the material is not standard stereo.
3. 03Check the commandthe preview area shows the ffmpeg arguments about to run, such as pan and channelsplit, and the split mode points out that two files will come out.
4. 04Start processingclick the button for the current operation and download the result when the job finishes; inputs such as WMA and AMR that cannot keep their format fall back to MP3.

Input range

the panel accepts audio only; swap, split and karaoke all rely on standard stereo, so mono material may fail or sound very poor.

How the vocal cut works

it is a straight left-minus-right subtraction, not a source-separation model; bass, snare and backing vocals that also sit in the centre can be thinned along with the voice.

Output format

it re-encodes in the source extension where possible; formats the current build cannot encode, such as WMA and AMR, come out as MP3.

What it does not do

it cannot set the left and right levels independently and cannot truly separate voice from backing; use “Volume” for overall gain and “Mix” to combine tracks.

The audio file is read and processed only locally in the browser by ffmpeg.wasm and is never uploaded; the first use of the workbench downloads about 31–32 MB of single-threaded ffmpeg core from the site's asset domain or jsDelivr, and that CDN request contains none of your audio.