Vocal Splitter — Separate Vocals from Instrumental, Locally
Drop in a song, get the vocals and the backing track as two separate files. Powered by Spleeter and Demucs running entirely in your browser — no upload, no account, and your unreleased music stays on your machine. Free to try — no subscription.
Split a Song Free →Two AI engines, one timeline
Source separation is a prediction problem: a network reads the song\'s spectrogram and estimates, for every moment and frequency, how much of the energy belongs to the voice. How heavy that network is decides the quality-speed trade-off — so this tool ships with two.
| Spleeter (default) | Demucs (WebGPU) | |
|---|---|---|
| Runs on | Any recent browser, phones included | Chrome / Edge with GPU (WebGPU) |
| Stems | 2 — vocals + instrumental | 2, or 4 — vocals / drums / bass / other |
| Max length | 30 min (desktop) · 3 min (mobile) | 8 min |
| Quality | Everyday good — light artifacts on dense mixes | Cleaner — fewer artifacts, better detail |
| Speed | Fast on CPU | Fast on GPU, heavy model |
Whichever engine you pick, both stems come out of a single pass — you never run the separation twice to get the other side.
Want the vocal side? →
The acapella workflow: remixes, sampling, cover practice and ear training. Read the acapella extractor guide.
Want the backing track? →
The karaoke workflow: turning any song into a singable instrumental, with quality tips for cleaner backing tracks.
Why run separation locally instead of on a site?
- Unreleased work stays unreleased — Uploading a demo to an unknown server is a leak waiting to happen. Local processing means your hard drive is the only copy — as it should be.
- No queue, no quota — Upload splitters meter your minutes and make you wait in line. Here the GPU/CPU you already own does the work, as many times as you like.
- Works offline — Once the page and model are cached, separation works without a network at all.
How to separate vocals from a song
- Drop your audio (MP3/WAV/FLAC/M4A/OGG) or video (MP4/MOV/MKV) onto the splitter.
- Choose Spleeter for speed and reach, or Demucs for quality if WebGPU is available.
- Run it — everything processes locally; you\'ll see progress, not an upload bar.
- Preview each stem, then download both in the format you want.
Frequently asked questions
Which engine should I choose — Spleeter or Demucs?
Spleeter runs on any recent browser (including phones) and is fast — use it for everyday jobs up to 30 minutes. Demucs needs a WebGPU-capable Chrome or Edge and handles tracks up to 8 minutes, but a newer, heavier model produces noticeably cleaner separation, with an optional four-way split. If the option appears in your toolbar, try it first.
What are the four stems in Demucs mode?
Vocals, drums, bass, and "other" (guitars, keys, synths — everything that isn't voice, drum or bass). The default two-stem mode gives you the classic vocals / instrumental pair.
Does the song get uploaded to a server?
No. The model downloads to your machine once and the separation runs entirely inside your browser tab. This matters for unreleased work: demos, session recordings and client material never leave your disk.
Can I split a video file, not just audio?
Yes. Drop an MP4, MOV or MKV — the audio track is decoded locally and separated, and the video is ignored. Useful for live performances and MVs.
What output formats are supported?
MP3, WAV, FLAC and OGG. WAV/FLAC for production work, MP3 for quick listening and karaoke apps.
Is the vocal splitter free?
It runs locally and is free to try — no sign-up, no watermark, no subscription (a one-time upgrade removes the daily quota).
Related guide
Guide ·
Spleeter vs Demucs: Which AI Stem-Splitter Sounds Better?
Deezer's Spleeter or Meta's Demucs — which AI model separates vocals better? Architecture, quality, speed, and when to use each, from a tool that ships both.