This free vocal remover splits any song into vocals and instrumental using a neural network that runs completely in your browser. Unlike server-based vocal removers, your music never leaves your device — there is nothing to upload, no queue, and no file-size pricing. The AI model (about 40 MB) downloads once and is kept on your device, so future splits skip the wait — and even if your connection drops mid-session, separation keeps working.
Common Use Cases
Make a karaoke track
Split the song, mute the vocals, and download the instrumental. Combine with Change Pitch to move the song into your vocal range — the pitch tool works far better on an instrumental than on a full mix.
Extract an acapella for remixing
Solo the vocals and download them as WAV. Producers use isolated vocals for remixes, mashups, and sampling.
Practice an instrument
Remove the original vocals to play or sing along with the band, or keep only the vocals to study a melody line.
Clean up speech over music
For videos where a voiceover competes with background music, separating the two lets you rebalance them instead of re-recording.
How to Remove Vocals from a Song Online
- Upload your song. The AI model downloads in the background the moment you open this page, so it is usually ready before your file is.
- Open Split Stems in the AI section and click Split Stems. Separation runs entirely on your device — the song is never uploaded.
- Listen to the result: mute the vocals for a karaoke track, or solo them for an acapella. Adjust each stem’s volume to taste.
- Download each stem as WAV or MP3 (or everything as one ZIP), or click Edit to load a stem into the editor for trimming, EQ, or pitch changes.
Common Misconceptions
Myth: Vocal removers just cancel the center channel
Reality: Old "karaoke" filters inverted one stereo channel against the other, which only worked for center-panned vocals and gutted the bass. Neural networks like this one learn what a voice actually sounds like and separate it even from mono or wide-panned mixes.
Myth: Online vocal removal requires uploading your song to a server
Reality: This tool runs the neural network in your browser with WebAssembly. The AI model is downloaded to you — your audio never goes the other way.
Common Problems and Fixes
The instrumental still contains faint vocal echoes
Reverb and delay tails of the voice are spread across the whole mix, and no separator can fully remove them. The result is usually clean enough for karaoke; heavily produced vocal effects leave more residue.
The vocals output sounds muffled or dull
The AI model analyzes frequencies up to 11 kHz, so isolated vocals lose some "air". Keep the "above 11 kHz" option enabled so the instrumental keeps its full brightness — that is where it matters most.
Separation is slow on my phone
Processing time depends on your device’s speed and the song length — on a recent computer a typical song takes about a minute; on phones expect a few minutes. Keep the screen on: the tool requests a wake lock, but switching apps can pause processing.
A cappella tracks or live recordings separate poorly
The model was trained on studio music with a clear vocal/instrumental balance. Live recordings with audience noise, or tracks where one instrument dominates, give less predictable results.
Frequently Asked Questions
Is this vocal remover really free?
Yes. Processing happens on your device, so there are no server costs to pass on — no trial limits, watermarks, or signup.
Is my music uploaded anywhere?
No. The AI model runs locally in your browser. Once the model has downloaded, you can even lose your connection while the page is open — separation still works.
What is the model download and how big is it?
The vocal remover’s AI model is about 40 MB. It downloads once in the background and is stored in your browser, so repeat visits skip the download entirely.
What audio formats are supported?
Everything the editor can open: MP3, WAV, FLAC, OGG, M4A/AAC and more. Each stem exports as 16-bit WAV or 192 kbps MP3, individually or as one ZIP.
How long can the song be?
Up to 30 minutes with the Fast model and 12 minutes with Studio quality, with a soft warning for long files. A typical 4-minute song separates in a minute or two with Fast.
Can I split drums and bass too?
Yes — switch to the Studio quality (4-stem) model in the tool’s model picker. It runs on your graphics chip and needs a browser with WebGPU (any current Chrome, Edge, Firefox or Safari). See the Stem Splitter page for details.