How to use this workflow
Check the device requirements
This tool requires a browser with usable WebGPU. Starting separation loads or checks an approximately 181 MB model plus runtime assets. There is no separate model-download button.
Choose a recording and start
Select a supported file no larger than 250 MB and no longer than six minutes, then select Split four stems. Available memory and browser storage can impose tighter limits. Try a short excerpt first.
Listen and download
Run separation, then use synchronized playback, mute and solo to inspect the stems. Download the individual WAV files or choose Drumless, Bassless, Instrumental or Acapella from the export choices.
What the four outputs contain
Vocals estimates singing and other vocal material. Drums estimates percussion, Bass estimates bass instruments, and Music contains the remaining instrumental material. These labels describe the model’s categories, not guarantees of clean isolation.
Swipe or scroll sideways to read all columns.
| Export | Intended content |
|---|---|
| Drums / Bass / Music / Vocals | One estimated stem per WAV file |
| Drumless | Bass + Music + Vocals |
| Bassless | Drums + Music + Vocals |
| Instrumental | Drums + Bass + Music |
| Acapella | Vocals |
Listen for leakage and missing detail
Cymbals can spill into other stems; reverberant vocals may remain in the instrumental mix; an instrument sharing the voice’s frequencies can be partially removed. Solo a stem to inspect the problem, then listen to the intended mix as well.
The tool converts audio to a 44.1 kHz stereo model input. Model files may be cached in browser storage, and processing can use temporary local buffers. There is no CPU fallback for this four-stem tool. It is not a cloud job that continues after the page is closed.
Common questions
Can I extract guitar as a separate stem?
There is no dedicated guitar stem in this model. Guitar usually shares the Music output with other instruments.
Are the stem downloads MP3 files?
The splitter’s download choices are WAV. Use the audio converter afterward if you need MP3 and accept the additional lossy encoding step.