Split vocals and music with one file drop
The vocal splitter above turns one mixed song into two clean tracks. The first is the vocal track: lead vocal, harmonies, ad-libs and any rap or spoken parts. The second is the music track: drums, bass, guitars, keys, synths and everything else. Each one is a full-length WAV file that lines up perfectly with the original, so you can drop both into an editor or a DAW and they stay in sync.
Behind the vocal splitter is HT-Demucs, an open AI model from Meta AI Research that separates music into vocals, drums, bass and other. For the two-track split we simply combine drums, bass and other into the music track. Because the three instrument stems are still there, you can rebalance the music track before you download it, for example to make the drums quieter under a spoken intro.
How to use the vocal splitter
- Drop a song into the box, or click it to pick a file from your computer.
- Wait while the vocal splitter runs. The first run downloads the AI model (about 170 MB) and caches it.
- Press play to check the split. Solo the vocals to hear only the voice; mute them to hear only the music.
- Click "Download vocals" and "Download music" to save the two tracks.
Vocal splitter outputs explained
| Track | Contents | Format |
|---|---|---|
| Vocals | All singing and speaking voices in the song | WAV, 44.1 kHz, 16-bit stereo |
| Music | Drums + bass + other instruments, mixed at their original levels | WAV, 44.1 kHz, 16-bit stereo |
| Optional: 4 stems | Vocals, drums, bass and other as separate files | ZIP of WAV files |
Played together, the vocal track and the music track add up to the original song. That makes the vocal splitter safe to use in a session: you lose nothing, you only gain control.
What people use a vocal splitter for
Mixing practice
Re-balance a famous song yourself. With vocals and music on separate faders you can hear how loud the vocal really sits in a professional mix, and practise EQ and compression on a real lead vocal.
Remixes and mashups
Put the vocal of one song over the music of another. Split both songs, match their tempo and key in your DAW, and combine them. The acapella extractor and the vocal remover pages are tuned for each half of that job.
Transcription and lyrics
Lyrics are much easier to hear with the music gone. Language learners and subtitle editors split a song to transcribe every word, then check timing against the vocal track.
Video and podcast editing
Lower only the singing under a voice-over while the beat carries on, or cut between the full song and the instrumental for a dramatic edit.
Tips for a better vocal split
- Feed the vocal splitter a high-quality source. Low-bitrate files have compression artifacts that the model cannot tell apart from the voice.
- Heavily processed vocals (vocoder, extreme Auto-Tune, distortion) may be partly assigned to the music track. Check them with solo before you download.
- Very quiet backing vocals sometimes stay in the music track. Turning the music track down a little in your mix hides that.
- If you need a lighter vocal in the final mix instead of none, set the vocal volume slider and use "Download mix as heard".
Vocal splitter vs stem splitter
A vocal splitter gives two tracks; a stem splitter gives four. They use the same AI here, so the quality of the vocal track is identical. Choose the two-track view when you only care about voice versus music, and the four-stem view when you also want to work with the drums or the bass on their own. You can switch between the two simply by soloing and muting in the mixer; nothing needs to be processed again.
Privacy and limits
The vocal splitter runs in your browser tab using your graphics card through WebGPU. Your audio is decoded and processed locally and is never uploaded. There are no accounts, no credits and no watermarks. The only practical limits are track length (10 minutes) and your computer's memory. Recent Chrome and Edge give the fastest results; other browsers fall back to slower CPU processing.