Karaoke makerBeta

Free karaoke maker

Drop a song and get a karaoke track: the music without the singer. Keep a little of the voice as a guide if you like. It runs on your device, so nothing is uploaded and there is nothing to pay.

  • Free
  • No sign-up
  • Runs in your browser
  • Nothing is uploaded

Loading the tool

How it works

  1. 01

    Drop the song you want to sing

    Audio or video from your device: MP3, WAV, FLAC, M4A, OGG, MP4 and more, up to 10 minutes (6 in HQ). The file never leaves your device.

  2. 02

    Let the AI take the singer out

    A neural network splits the song into vocals and music in your browser. Fast (Demucs v4) is an 85 MB download. HQ (BS-RoFormer) is 353 MB and cleaner. Your browser keeps the model for next time.

  3. 03

    Set the guide vocal, then save

    Play the karaoke track and flip to the original to compare. Slide the guide vocal up if you want to hear the tune, then download the WAV.

How to make a karaoke track

Making a karaoke version takes three moves. Drop the song, wait while the split runs, then download the track. Whatever file your browser can play works: an MP3 from your library, a WAV from a session, even the sound of a video file. A karaoke track is the song with the voice taken out and everything else left where it was: the drums, the bass, the guitars, the keys, the strings and the effects.

Under the hood, this karaoke version maker splits the song into its parts with a neural network, then adds every part back together except the vocals. That is the whole trick, and you can hear it before you save: switch between Karaoke and Original to check that the track is the one you want to sing over.

A karaoke track is audio only. For the words, keep the lyrics on your phone, on a printed sheet or in any lyrics display you like, and press play on the track.

Singing with a guide vocal

A clean karaoke track is what you want when you know the song. When you do not, a bare backing track can leave you guessing where the melody goes. The guide vocal slider keeps a little of the original singer under the music, like a pencil sketch of the tune. It starts at Off, and it moves in steps of 2 dB, from very quiet up to -6 dB, which is about half the original level.

A good way to use it: learn the song with the guide a little higher, then lower it a step each run until you can sing without it. For a performance, set it to Off. The slider changes what you hear right away, and the file you download has exactly that level in it. A guide vocal in the file shows in its name, for example Karaoke, guide -12 dB, so you never mix up the versions.

Keep in mind that the guide is the original voice, with its words and its phrasing. It is there to point at the notes, not to hide a missed one, so keep it low if you want the track to sound like karaoke.

What is in the track, and what is not

The model hears every voice as a vocal. Lead and backing voices, harmonies and ad-libs all leave together, so a song with a big choir becomes a very empty track. If you want some of that back, the guide vocal brings back the whole vocal part, backing voices included, at the level you pick.

We do not add lyrics or timing, and we do not change the key or the tempo. If the song sits too high or too low for you, change its key in your DAW or a pitch tool once you have the karaoke track. The track comes out at 44.1 kHz stereo, whatever the sample rate of your file was.

Tips for a cleaner karaoke track

Start from the best file you have. A WAV or a high-bitrate file separates more cleanly than a small MP3, because compression blurs the detail the model relies on. If a trace of the voice stays in the track, run the song again in HQ: it is a newer model that separates more cleanly on public benchmarks.

Listen to the whole track once before the night. Dense mixes, heavy reverb on the voice and instruments that sound like a voice, such as a synth lead or a vocal chop, can leave a faint ghost of the singer. Near the end of a song, a long reverb tail can carry on for a second or two after the voice has stopped.

Keep the volume sensible. If any part of the sum would pass full scale, we lower every file by the same small amount and say so under the player, so the track never clips. Turn the track up on the speaker afterwards.

Parties, rehearsals and covers

A karaoke version helps well beyond the party. Singers rehearse with the real arrangement instead of a copy, choirs practise a part against the recording, and cover artists record their own take over the original backing. Teachers hand students a track for the song they picked.

Once you have sung over the track, the next step is the Vocal + beat mixer: it takes your karaoke track as the beat, you add your recording, and you hear the two mixed and mastered.

Nothing to upload, nothing to pay

The work happens on your device, so your song is never uploaded and no server bill forces a limit on you. There is no sign-up, no credit counter and no daily cap, only a length limit of 10 minutes per song, or 6 in HQ. The tool is in beta: it works, it is new and it is getting better. Tell us what you think at info@obra.art.

What you get

  • Takes the singer out and keeps the whole band: drums, bass and every other instrument
  • Guide vocal slider, from off up to -6 dB, to keep a little of the voice under the music
  • A neural network that runs in your browser: Demucs v4 in Fast, BS-RoFormer in HQ. Your song is not uploaded
  • Hear the karaoke track and the original side by side before you save
  • Lossless WAV at 44.1 kHz, 24-bit or 16-bit. The file name says when a guide vocal is in it
  • One click to take the track into OBRA's Vocal + beat mixer and add your own voice
  • No sign-up, no credits, no daily limit

A karaoke version from this page, from a service that uploads, or ready-made

OBRA karaoke makerOnline karaoke makers that uploadReady-made karaoke tracks
How the singer comes outA neural network separates the voice from the music of your own fileMostly a neural network on their serversThe band is usually played again in a studio, so there is nothing to separate
Which songsAny audio or video file you haveAny file you uploadOnly what the catalogue has
Your songStays on your deviceUploaded to a serverNothing of yours is involved
Guide vocalA slider from off to -6 dBVaries by serviceRare
Lyrics on screenNo, audio onlyVaries by serviceOften included
Sign-up and costNone. Free, no creditsOften free with limits, or paid plansFree or paid, depending on where you get it
SoundThe original recording, minus the voiceThe original recording, minus the voiceUsually a copy of the arrangement, not the recording

Based on public information in November 2026. Services change their plans often, so check them before you decide. We are not affiliated with any service named here.

Frequently asked questions

Is this karaoke maker really free?

Yes. There is no account, no card, no credits and no daily limit. It is in beta.

Does my song get uploaded anywhere?

No. The split runs in your browser on your own device, so the song never leaves it. The first time, your browser downloads the AI model (85 MB for Fast, 353 MB for HQ) and the code that runs it from public hosts. Your audio is not part of that.

Can I make a karaoke version of any song?

Of any audio or video file on your device, up to 10 minutes in Fast and 6 in HQ. We do not take links to streaming sites or videos: make your own file first, from music you own or may use.

Does it remove the backing vocals too?

Yes. The model treats every voice as vocals, lead and backing alike, so all of them come out of the karaoke track. The guide vocal brings the whole vocal part back at a low level, backing voices included. There is no setting to keep the backing voices and drop only the lead.

What does the guide vocal do?

It mixes a little of the original singer back in under the music so you can follow the melody. Off gives a clean karaoke track. Each step is 2 dB louder, and the top is -6 dB. The WAV you download has the level you hear, and the file name says so.

Does it add lyrics or change the key?

No. We make the audio only: no words on screen, no timing, no key or tempo change. Use the track in a lyrics display, a media player or your DAW, and change the key there if the song does not suit your range.

Should I use Fast or HQ for karaoke?

Fast is an 85 MB download that runs on a graphics card or your processor and takes songs up to 10 minutes. HQ is a 353 MB download for a computer with WebGPU, takes songs up to 6 minutes, and leaves less of the voice behind, but it is much slower. On a MacBook Pro with an M1 Pro chip, a 3 minute song took about 45 seconds with Fast on its graphics card and about 7.5 minutes with HQ. On a phone, Fast can work with limits and HQ is not offered.

Can I use the karaoke track at a party or in a video?

Singing at home or with friends is ordinary use. The song still belongs to the people who made it, so to perform it in public, post a video with it or sell anything made from it, you need permission from the rights holders. This is general information, not legal advice.

Learn more, or get a person on it

Read the mastering guides, comparisons and data, or send your tracks to an engineer with OBRA ULTRA.

Happy with the result?

Master it on OBRA Audio Mastering: three masters measured in LUFS, free to listen, in your browser.