Three simple steps
How to separate a song into stems
- 01
Choose an audio file from your device. The file stays in your browser and is not uploaded to ToolYogi.
- 02
Start separation. On first use your browser downloads the AI model, then uses your device’s GPU when WebGPU is available or falls back to WebAssembly.
- 03
Preview vocals, drums, bass and other, then download the WAV stems you want. Refreshing or closing the page clears the session.
What stems does this tool create?
The current browser model separates a track into four parts: vocals, drums, bass and other. The other stem contains the remaining musical content that is not classified as vocals, drums or bass.
Does my song get uploaded?
No. The selected audio is decoded and processed in your browser. ToolYogi does not provide an upload endpoint or cloud history for this stem-separation tool. The AI model itself is downloaded separately as a static file.
Why can stem separation take a while?
AI source separation is much heavier than ordinary audio analysis. Speed depends on your browser, available memory and hardware. WebGPU-capable devices can be substantially faster; WebAssembly fallback may be slow, especially for long songs.
Why are the stems not perfectly clean?
Source separation is an estimate. Sounds that overlap in frequency or timing can leak between stems, and results vary by recording, arrangement and mix. The output is useful for practice, remix preparation and experimentation, but it is not identical to original studio multitracks.
A little more clarity
Frequently asked questions
Is the Stem Separator free?
Yes. The current ToolYogi Stem Separator runs on your device and does not require an account.
Does ToolYogi store my song or stems?
No. This tool does not upload your audio or stems to ToolYogi servers and does not create a cloud history. Generated download URLs exist only in the current browser session.
What stems will I get?
The current model creates vocals, drums, bass and other stems.
Why does the first run download a large file?
The AI model is roughly 172 MB. Your browser needs the model to perform separation locally. Browser caching may avoid downloading it again on later visits.
Which browser works best?
A current browser with WebGPU is preferred. If WebGPU is unavailable, the tool can attempt WebAssembly processing, which may be considerably slower.
Can I separate any copyrighted song?
Only process audio that you own or have permission or another lawful basis to use. ToolYogi provides a local processing utility and does not grant rights to third-party recordings.
Open-source local processing
Browser stem separation is based on the MIT-licensed demucs-web project and the HTDemucs model architecture. ToolYogi keeps the processing flow local to your browser and does not provide a server upload endpoint for this tool.