Upload a recording and get its parts back as separate tracks. You can ask for the full split (vocals, drums, bass, and the remaining instruments as four tracks) or isolate one stem and keep everything else mixed into a single "rest" track. The tool runs a neural source separation model that works at the spectral level, so each instrument family comes out cleanly even from recordings where everything overlaps in the mix.

How to use it

  1. Drag and drop an audio or video file, or click the drop zone to browse (MP3, WAV, FLAC, OGG, M4A, WebM, and MP4 are all accepted). The file must be under 50 MB. There is no length limit on the file itself; you pick the section to process in the next step.
  2. The full waveform appears. If your file is longer than four minutes, two draggable markers (green for start, red for end) let you pick the section you want to split, up to four minutes wide. You can drag the markers, click anywhere on the waveform to move the start to that point, or use the small « ‹ › » buttons on either side to nudge the start by one or ten seconds. Press the small Preview button to listen to your selection before committing.
  3. Choose what you want to split. All stems returns four separate tracks (vocals, drums, bass, other). Vocals + rest, Drums + rest, Bass + rest, or Other + rest isolates the chosen stem and bundles everything else into a single "rest" track, useful when you only need one part and don't want to deal with four downloads.
  4. Click Analyze & Split. The tool tells you how many short ads are needed for your selection (one per minute of audio, up to four), then plays them. The upload begins in the background while the first ad plays, so you don't wait on it separately.
  5. Once the ads finish, the resulting tracks appear as cards. Each one has its own play button, volume slider, and waveform you can click to scrub. Tick the cards you want, then use Download per row, Download all to grab everything, or Open ticked with… to send the selected tracks straight into another tool, including the Audio Merger when you have multiple ticked tracks to work with side by side.

What each stem contains

  • Vocals: lead and backing vocal performances, including harmonies.
  • Drums: kick, snare, hats, cymbals, and percussive hits.
  • Bass: the bass guitar or synth bass that carries the low end.
  • Other: everything else, typically rhythm guitars, keys, pads, horns, strings, and similar layered parts.
  • Rest: only appears when you pick one of the single-stem options. It contains every part of the song that is not the stem you isolated, mixed back into a single track.

FAQ

Is there a length limit on the file I upload? No, you can upload any length. Each split runs on up to four minutes of audio at a time, so for longer files the tool shows draggable start and end markers on the waveform and you choose which section to process. Anything outside that window is left untouched.

Why pick "Vocals + rest" instead of "All stems"? If you only need one part (say the vocal line to practice harmonies over, or just the drums for transcription), the single-stem options are faster to download and easier to handle than four separate tracks. The audio quality of the isolated stem is the same in both modes.

How long does processing take? Processing starts after the last short ad finishes and usually takes between a few seconds and about a minute, depending on the length of your selection and current load. The page shows a status while it runs.

Why does the tool show short ads? Running the separation model uses a GPU accelerator that costs real money for every file. Short ads make the tool free for everyone without subscriptions or accounts. You only watch one short ad per minute of audio you choose to split, so a thirty-second selection only asks for one short ad, and a four-minute selection asks for four.

Does the quality depend on the source? Yes. Clean studio recordings produce the cleanest stems. Noisy live recordings, low bitrates, and heavy effects on any single instrument make separation harder, but you will still get usable results in most cases.

What sample rate do the stems come back at? Each stem is returned as a 16-bit stereo WAV at 44.1 kHz, the model's native processing rate, so the tracks are drop-in ready for any DAW or mixing app.