Free AI Vocal Remover and Karaoke Maker: Remove Vocals From Songs
About this vocal remover
A free AI vocal remover that never uploads your song
This vocal remover takes the singer out of a song and leaves the music standing, and it does the work inside your own browser, on the phone or computer in front of you.
Anyone who has tried to practise an old film song knows why that matters. The karaoke version either does not exist, or somebody has uploaded one in the wrong key with a channel logo floating over it.
Big online removers help, up to a point. On its free plan, vocalremover.com converts only 30 seconds of a song. X-Minus allows 15 minutes of audio a day. Most of them want the file on their own computers first, and Tunebat says plainly that your song is uploaded and processed on its backend.
Here nothing goes up. The only things that travel are the AI models and the engine that runs them, and they travel the other way: down to your device, once.
Express or Studio?
The page carries two different machines, and the Mode buttons under the player choose between them.
Express runs Spleeter, which the research team at Deezer released as open source in 2019. It reads a song as a picture of frequencies moving through time and decides, patch by patch, how much of each patch belongs to the voice. Light and quick, it is often all you need for everyday practice.
Studio runs UVR-MDX-NET Inst HQ 3, a model trained by Kimberley Jensen for the Ultimate Vocal Remover project. It is a much heavier model, and it was trained the other way round. Instead of hunting for the voice, it hands back the instrumental itself. On most songs the karaoke comes out cleaner, with less of the singer left hanging in the loud passages.
| Express | Studio | |
|---|---|---|
| AI model | Spleeter, by Deezer | UVR-MDX-NET Inst HQ 3 |
| First-time download | about 39 MB | about 67 MB |
| Time on a phone | the quicker of the two | a few minutes to over an hour |
| Reach for it when | you want to start singing soon | the track has to be as clean as it can get |
My advice is to start with Express. If the karaoke is fine for practice, you are finished. Should a faint voice still float under the chorus, give the same song to Studio, put the phone down on the table and let it work.
How to remove vocals from a song
- Tap the empty timeline under Add a song to start and choose an MP3 or WAV file. On a computer you can drop the file onto it instead.
- Under Mode, pick Express or Studio.
- Tap ✨. The first time, the model downloads; after that the browser keeps it and opens it straight away.
- Press Play when it finishes. The 🎧 button flips between the original and the karaoke at the same moment in the song.
- Tap Save to keep the karaoke as a WAV file.
Before you save, listen to one verse and one chorus with 🎧. Ten seconds of switching back and forth will tell you more than anything written on this page. The timeline helps too. It draws the karaoke over a faint copy of the original, so the stretches where only the singer could be heard visibly shrink.
The wait, honestly
Express is the quick one. Studio is not, and I would rather say so here than let a slow progress bar say it for me. On a phone, one song through Studio can take anything from a few minutes to more than an hour.
The reason is plain arithmetic. Studio listens to the song in pieces of a little under six seconds, each one spread across 3,072 frequency bands in both channels, so a four-minute song means about forty rounds of heavy work. And a web page like this one gets just one core of the phone's processor to do it with. A laptop gets through the same song sooner.
While it works, the screen stays awake by itself, and after the first piece the line under the timeline shows roughly how much time is left. Keep the page in front, though. Switch to another app and the browser may pause the page, and the work waits with it. If a phone runs out of memory halfway, the page says so the next time it opens; Express is the lighter choice for that phone.
A karaoke maker for songs nobody made a track of
Most people who go looking for a karaoke maker are not planning a party. One is a singer doing riyaz on a bhajan whose track was never released. Another is a student with one song to prepare for the school function, or someone recording a cover in the evening after work, or a music teacher. The song exists. Its accompaniment, on its own, does not.
Teachers find a second use. With the voice gone, you can hear what the harmonium is really doing under a line you thought you knew by heart. It is a little like seeing a familiar room with the furniture carried out.
Some karaoke makers online, Youka for one, build videos with the lyrics scrolling across the screen. This page does something smaller. You get the audio, in the song's own key. It does not change the pitch, so if the original sits too high for your voice, the track will too. Keep the lyrics on paper, or on a second screen.
What the AI leaves behind
No vocal remover is perfect, and the ones that promise perfection are selling something.
- A chorus or backing vocals are harder to remove than a single lead voice. Traces often remain, more with Express than with Studio.
- Heavy reverb on the voice can leave a faint ghost of it behind, like someone singing in the next room.
- Instruments that sing, a sarangi or a flute carrying the tune, sometimes lose a little along with the voice.
- A worn file gives the AI less to work with. An MP3 saved at 128 kbps years ago tends to come out slightly watery.
Old mono recordings are the pleasant surprise. The trick free vocal removers relied on for years, cancelling whatever sits in the middle of the stereo picture, does nothing at all to a mono song. This page began with that trick, and it was thrown out after a few real songs. Both models here listen to the sound itself, so a mono recording from the 1960s can be cleaned too.
What you get back
A WAV file, 44.1 kHz stereo, about 10 MB for every minute. The size surprises people who started with a 4 MB MP3. Saving it as MP3 again, though, would squeeze the sound a second time on top of the squeezing the original already had.
Need only one verse? Trim the karaoke in the free online audio editor. The online audio joiner will stitch two tracks into a medley, and an old recording with tape hiss can go through the background noise remover on its Light setting.
Whose work this runs on
Express uses Spleeter from Deezer. Studio uses a model trained by Kimberley Jensen and shared through Ultimate Vocal Remover, a free project whose developers ask one thing of anyone who uses their models: credit them. Gladly. Both are released under the MIT licence.
One plain word about songs. A karaoke track made here is for your own practice. Uploading it to YouTube or selling it needs permission from whoever owns the song, exactly as the original would.
If a song came out badly, write to me with its name and the mode you used; those are the cases I learn most from. The other tools are on the Audio Tools page, and everything else lives on the Critical Monk home page.
Questions people ask
Is this vocal remover really free?
Yes. There is no account, no daily limit on minutes and no watermark on the karaoke. What it costs is your device's time, and with Studio that can be a lot of it.
Is my song uploaded anywhere?
No. Your song is read and processed inside your browser. The only downloads are the AI models and the engine that runs them, once, the first time you use each mode.
Which is better, Express or Studio?
Studio usually gives the cleaner karaoke and Express the quicker one. Try Express first, and move to Studio only if the voice still shows through.
Why is Studio so slow on my phone?
It runs a much bigger model over the whole song in six-second pieces, and a web page gets only one core of the phone's processor. Keep the page open; the screen stays on by itself while it works.
Can I get only the vocals, without the music?
Not here. This page keeps the music and lets the voice go.
Which files can I use?
MP3 and WAV. An M4A, an AAC or the sound from a video has to be converted to one of those first.
Will it change the key of the song?
No. The karaoke stays in the key the song was recorded in.