Vocal remover & stem splitter without upload
Split a song into its stems – vocals, drums, bass and the other instruments. Mute the vocals for a karaoke version, listen to the voice alone, or practise along without the bass. There are two levels: the best separation with HT Demucs, or a fast one with Spleeter that is quick even without a graphics card. Everything runs on your device; the song is not uploaded anywhere.
Pick a song on the left – it is split into vocals and instruments.
While it is working you hear the original. Afterwards you set every stem separately – “Mix as WAV” saves exactly what you hear.
How it works
Choose a song
Drop an audio or video file on the field or pick it. The first time, the browser downloads the model (57 MB, 20 MB in fast mode); separation starts right away.
Mix the stems
When it is done you set every stem separately – vocals, drums, bass and other, or vocals and accompaniment in fast mode – or pick a preset such as “Karaoke”. All stems play in sync.
Save
Save your mix as WAV or Opus or download single stems – just the vocals, say, or just the instrumental.
Frequently asked questions
Is my song uploaded?+
No. The song is separated in your browser and does not leave your device. The AI model and the runtime are served by mytoolio itself – no other provider is involved.
How good is the separation?+
The “Best” level uses HT Demucs, one of the best freely available models: with studio recordings the vocals usually come apart cleanly. Some residue is normal all the same – a trace of the voice in the instrumental, some reverb or quiet instruments in the vocal stem. The “Medium” and “Fast” levels (Spleeter, four or two stems) separate more coarsely: more residue, and above 11 kHz it can only roughly tell vocals and instruments apart. Live recordings, very old or heavily compressed tracks and choirs are harder at every level.
How long does it take?+
“Best” level: with graphics-card support (WebGPU in Chrome and Edge) about a sixth of the song’s length – a three-and-a-half-minute track takes roughly 35 seconds; without WebGPU about four times as long as the song. “Fast” level: about a quarter of the song’s length even without a graphics card (longer on a phone), which is why “Fast” is preselected when there is no WebGPU. The “6 stems” level takes a little longer than “Best”, and very long without a graphics card. You can listen to the original meanwhile.
Which files work?+
Anything your browser can play: MP3, WAV, M4A/AAC, OGG/Opus, FLAC, WebM and the audio track of MP4 or MOV videos – up to 10 minutes long (6 on a phone). The results are stereo WAV files at 44.1 kHz.
May I use the separated stems?+
For yourself – to practise, sing along or analyse – yes. Separation does not change the rights to the music, though: anyone who wants to publish or reuse an instrumental or a vocal stem from someone else’s song needs permission from the rights holders.
Does it work offline?+
Yes. After the first download the model and the runtime stay in the browser cache; after that the tool also works without an internet connection.
Can I save stems smaller than WAV?+
Can I get guitar and piano on their own?+
Where do key and tempo come from?+
Separation with Hybrid Transformer Demucs (Meta, MIT licence) or Spleeter (Deezer, MIT licence) on onnxruntime-web – entirely on your device. The stems are an estimate by the AI, not the original recordings of the individual instruments. Please mind the copyright of the songs you process.