Split a drum loop into kick, snare and cymbals
Pull a drum loop or break apart into kick, snare, cymbals and the rest in your browser, then save each stem as a 24-bit WAV. Free, nothing uploaded.
Drop your audio file here
It opens in seance in this tab. The file stays on your computer.
- Opens
- WAVMP3FLACOGG
- Saves
- WAV
- Runs in
- Your browser
How to split a loop#
- Drop the audio file on the box above, or open seance and press Load audio. WAV, MP3, FLAC and OGG all work.
- Open Stems, leave Split the drum kit ticked and press Separate. To split only part of the file, drag across the waveform first.
- The rack fills with four stems: harmonic, kick, snare and cymbals. Each has its own level, solo and mute, and they play in time with each other.
- Save any stem as a 24-bit WAV, or open it as the working audio to slice or measure it.
Without the tick you get two stems instead: harmonic and percussive.
How good the stems are#
seance separates with signal processing, not a trained model: it splits what holds steady over time from what spikes across the spectrum, then sorts the spikes into kick, snare and cymbals by their sound. Every build measures the result on a synthetic groove whose true parts are known. Higher is cleaner:
| Stem | Quality (SI-SDR) |
|---|---|
| Snare | 11.6 dB |
| Cymbals | 8.9 dB |
| Kick | 7.0 dB |
| Harmonic | 4.1 dB |
In practice: the snare and the kick come out usable on their own, and the cymbals are good in a mix but not clean when soloed. Low notes that play together with the kick can end up in the kick stem. It works best on drum loops and breaks; on a full song, the percussive stem carries more than the drums.
What stays on your computer#
Everything. The file is decoded by your browser, the separation runs in a background thread of the page, and the WAVs are made there too. A long file takes seconds to minutes, and the page stays usable while it works.
Questions#
Can it remove vocals from a song?#
No. seance splits drums from everything else and the kit into its pieces. It has no vocal stem.
Why not use an AI model?#
A model would be a download of tens to hundreds of megabytes before anything happens, and its quality would be hard to check. This method is small, runs anywhere, and its numbers are measured on every build.
What format are the stems?#
24-bit WAV at the sample rate of your file.
Is my audio uploaded?#
No. Decoding, separation and export all happen in the page.