Before you begin
  • Use a recording you are allowed to process.
  • Start with 30–60 seconds before trying a long file.
  • Stemix currently supports up to 256 MB and 10 minutes per file.

Start with a question, and a short file

Decide what you want to hear differently. “I want the singer quieter while I practise” is a useful target; “I want perfect studio stems” is not a reliable expectation. Use a short excerpt with both a quiet passage and a busy section. You can judge the result before spending time on the whole recording.

Open Stemix in a current Chrome or Edge browser with WebGPU and hardware acceleration. Pick a local audio file or drag it onto the upload area. Selecting a file does not upload it. The file remains in your browser, and you can remove it from the queue before starting.

Choose the smallest useful model

For vocals and a backing track, choose 2 stems. Choose 4 when you need separate estimates for vocals, drums, bass and other instruments. The 5-stem option adds a piano estimate. More stems do not guarantee a cleaner vocal track; these are different model choices, not a quality slider.

Understand the first download

The download dialog lists the selected model and required components before you agree. These are processing resources, not your audio. They are cached in this browser for later use. Changing model or clearing site data can require another download. A private browsing session may not keep that cache after it closes.

Listen in context

  1. Play the completed result with all tracks audible. Check that the file and duration are the ones you intended.
  2. Mute the vocal track or lower its volume. Listen through a busy chorus as well as a quieter section.
  3. Listen to the isolated track too. A stem that sounds imperfect alone can still be useful in a practice mix.
  4. Lower other tracks if the combined mix sounds harsh. Adding several estimated stems can produce peaks above the original mix.

Save a track or a mix

Choose WAV when you plan to edit further, or MP3 for a smaller listening copy. The save menu offers each stem and Mix. Mix uses the current levels and muted tracks; an individual stem download saves that track. With several processed files, “Save all” packages one mix per file into a ZIP.

Completed results appear in local history and can be reopened in the same browser. That is not a cloud backup. Export anything you want to keep before clearing browser data, changing browser profiles or moving to another device.

This guide describes the current product controls and implementation. Separation results depend on the recording, device and model; examples are not quality guarantees.

Further reading: Spleeter · Web Audio API · How the tools work