01
How to remove vocals in 3 steps
- 1
Upload your track
Open Stem Separation and sign in with Google before submitting audio. Upload an MP3, WAV, FLAC, AAC, OGG or M4A file up to 30 MB. Your upload is private and is not published.
- 2
Let the AI split the track
The model listens to the full mix and rebuilds it as separate layers. Most songs finish in about a minute; longer or denser tracks take a little more.
- 3
Preview and export
Play the instrumental against the original before you download. If the balance is not what you wanted, you can pull the individual drum, bass and instrument layers instead of the single backing track.

02
What you can do with the instrumental

Sing over it
A backing track in the original key and tempo, ready for karaoke, live sets or a cover you record at home.

Record a cover
Keep the arrangement you love and put your own voice on top, without hunting for a studio-made instrumental that may not exist.
Practise your part
Mute the vocal to hear the arrangement underneath, or keep only the drums to practise timing against the real recording.
Sample and rework
Lift a clean drum or bass layer to build something new, then finish the track with AI generation.
03
When it does not work well
Vocal separation is very good, not magic. These are the cases where you should expect audible artefacts — knowing them up front saves you from blaming the wrong thing.
Live recordings
When the vocal bleeds into every room mic, there is no clean vocal to lift out. Expect a thin instrumental with traces of the singer left in.
Heavy reverb or delay
The model removes the voice but the tail of its reverb belongs to the room, not the singer. You will often hear a faint ghost of the vocal decaying after each phrase.
Old mono recordings
Modern separation leans on stereo placement as a cue. Pre-1960s mono masters give it far less to work with, and results are noticeably rougher.
Dense overlapping arrangements
Where a distorted guitar or a lead synth occupies the same frequency range as the voice, some of it may be removed along with the vocal, or some vocal may survive inside it.
04
Supported formats and limits
- Input formats
- MP3, WAV, FLAC, AAC, OGG, M4A
- Maximum file size
- 30 MB
- Typical processing time
- About 1 minute
- Output
- Lossless export, original key and tempo preserved
- Account required
- Reading this guide: no account needed. Processing audio: sign-in required; new accounts include free credits.
05
Vocal remover or stem separation — which do you need?
Both run on the same model. The difference is how much of the mix you want back.
Vocal remover
Gives you two things: the voice on its own, and everything else as one instrumental. Pick this when you want a backing track to sing over.
Stem separation
Breaks the mix into vocals, drums, bass and other instruments as separate files. Pick this when you are remixing or producing and need control over each layer.
Open the stem separator06
Frequently asked questions
Do I need an account to remove vocals?
You can read this guide without an account. To submit audio for separation, sign in with Google. Processing runs on your account and saves the result to your library. New accounts include free credits.
Is it really free?
New accounts come with free credits, and a vocals-and-instrumental split costs 10. That is enough to try several tracks before deciding whether to buy more.
Will the instrumental keep the original key and tempo?
Yes. Separation does not re-pitch or time-stretch anything, so the backing track lines up exactly with the original recording.
Why can I still hear a little of the singer?
Usually reverb. The model removes the dry voice, but the reverb tail is part of the room sound and is much harder to isolate. Studio recordings with a dry vocal come out cleanest.
Can I get the vocal on its own instead?
Yes. The same run produces the isolated vocal alongside the instrumental, so you can take an a cappella from any track you upload.
Is it legal to use the instrumental?
Separation does not change who owns the recording. Personal practice and private use are generally fine; publishing or monetising a cover normally needs a licence from the rights holder.
Turn that instrumental into something of your own
You have the backing track. Write over it, or let the AI build a full song from a single line of description.
Start creating