Start with a clean source
Upload a WAV, MP3, or M4A, then check the track profile. Shorter files and unclipped mixes are kinder to your browser.
AUDIO WORKBENCH
Create practical stem previews for remixing, practice, and production. Your source audio stays on this device.
01 / SOURCE
Drop an audio file here
No track selected yet.
02 / DESTINATION
Long tracks can take a few minutes. Keep this tab open while audio is being processed.
02A / QUALITY
Balanced uses smooth multi-stage filters and safe headroom for a practical browser render.
These are signal-processing estimates. A trained ML separator or original multitrack session is still the quality ceiling.
02B / SEPARATION MODE
Demucs can produce much cleaner vocals, drums, bass, and other stems, but it needs a server or GPU worker connected to this app.
Not connected yet. Enter your worker URL to enable true ML separation.
Worker response: return outputs with kind and URL, or a job ID/status URL that eventually returns those outputs.
Privacy: this mode sends the track to your configured worker. The app does not control that server's retention.
Local mode keeps your source audio in this browser.
02B / INSTRUMENT MAP
Scan a short excerpt to find likely keys, guitar, strings, and synth parts. Choose any detected parts to create focused practice stems.
No instrument scan yet.
Preparing your audio…
SAFETY CHECK
A quick check catches missing audio support before you spend time rendering.
03 / OUTPUTS
04 / SAMPLE LAB
Turn a detected instrument or any finished stem into a short, single-use sample. The app finds a lively moment, trims it, fades the edges, and exports a clean WAV.
CLEANER RESULTS