The best free vocal removers, compared honestly
Updated · 7 min read
Short answer: for a quick split with no account, vocalremover.org; for practice apps on your phone, Moises; for the most instrument types, LALAL.AI; for full control on a computer, Ultimate Vocal Remover or Demucs; and if you don't want to upload your song at all, Extractune, which splits it in your browser. We make Extractune, so we've put our own measurements here, including a test against vocalremover.org on one song scored against the artist's own stems, plus screenshots and the places where another tool is the better pick, so you can judge for yourself.
How to choose
Most free vocal removers use the same kind of AI model, so the difference is less the sound and more everything around it: whether your song is uploaded to a server, whether you need an account, what the free option includes, and what you can do with the parts afterwards.
The facts about vocalremover.org, Moises and LALAL.AI come from their own sites, read on 6 November 2026; those about Ultimate Vocal Remover and Demucs were checked in September 2026. Free limits and prices change often, so each comes with its date: check the site before you rely on one. We tested vocalremover.org against Extractune ourselves. Moises and LALAL.AI need an account, and we don't make accounts for tests, so they get their own sites' facts only.
Extractune: best if you don't want to upload your song
Extractune runs HTDemucs, the default model of Meta's open-source Demucs, inside your browser. The song or video never leaves your device, there's no account, and it's free with no monthly count. It splits voice, drums, bass and everything else, up to 6 minutes at a time where the browser can use the graphics chip (5 on devices with less memory, 2 without a usable graphics chip), and starts from a video as easily as from a song.
The split happens in a studio: a fader for each part, slow the mix down to 50% without changing the pitch, change the key, and save the video back with only the voice. The catch: the first split downloads the 90MB engine once, and a phone takes a while, as the measurements below show. There are no apps to install, and no extra instrument types beyond the four parts.
How fast it splits, measured on real devices
Because the split runs on your own device, its speed depends on that device. Every finished split on Extractune reports how fast it ran and on what kind of device (never the song or its name), so here are the real numbers rather than a promise.
| Device | Splits measured | A 2-minute clip | A 3-minute song |
|---|---|---|---|
| Computer with a graphics chip | 190 | under a minute | about 1 min |
| Phone with a graphics chip | 65 | about 3 min | about 4½ min |
| Computer without one | 18 | about 3 min | Over the 2-minute limit |
| Phone without one | 15 | about 7 min | Over the 2-minute limit |
How separating on your device works
The first time you split, your browser downloads the model once (about 90MB) and keeps it. It's saved in pieces as it arrives, so a dropped connection carries on where it stopped instead of starting again. After that, every split runs on your own graphics chip, or on the processor where the browser can't use it.
The model separates a song into four parts: the vocals, the drums, the bass, and everything else. Extractune lays the four parts on one timeline, with one-tap mixes (an acapella, an instrumental, a drumless track and more) and a fader, mute, solo and download for each. Long songs are split in 30-second batches, and you can listen to the finished part while the rest works.
Why the upload matters
An upload means your file sits on someone else's server, at least for a while, under their policy. For a demo you haven't released, a private recording, or a video with people in it, you may not want that. It's also slow on a phone: a big video has to upload before anything happens.
Splitting on the device turns that around: nothing leaves, but the work lands on your phone or computer, which is why the speed table above matters. If you split a lot of long songs on an older phone, a tool that splits on its servers can be quicker.
vocalremover.org: quick, with no account
A free set of audio tools in the browser that needs no sign-up, and a common first stop for removing vocals. Your file is uploaded and split on its servers. Its vocal remover gives two tracks, the vocals and a karaoke instrumental; a separate Splitter gives vocals, bass, drums and the rest, and other tools change the pitch or find the key and BPM.
Because it needs no account, it's the one we could test side by side. On 6 November 2026 we split "Just Lucky, I Guess" by Mr_Yesterday, a CC BY 4.0 song whose artist published its stems, with both tools on one computer, and scored the downloads against those stems. Extractune's parts came out cleaner, the times were close, and only vocalremover.org uploaded the song. One song on one computer is a small test, so the comparison page has the full method.
Pick it for a quick vocals-and-music split when uploading the file is fine.
The full test: method, plain SDR and the song
| Tool | Vocals | Instrumental | Time for the 3:18 song |
|---|---|---|---|
| Extractune | 10.76 dB | 8.31 dB | About 27 s, nothing uploaded |
| vocalremover.org | 9.79 dB | 6.38 dB | About 24 s, upload included |
| The song itself, not split | 0.64 dB | −2.06 dB | No split |
Moises: best for practice apps
Moises has a web app plus apps for desktop, iPhone, iPad and Android, and adds practice tools such as a chord finder, a speed changer and lyric transcription. Songs are uploaded to it, and it needs an account, so we haven't tested it.
What its own site says, as of 6 November 2026: the Free plan allows 5 uploads a month, with files up to 5 minutes long, split into 2 parts (vocals and instrumental) or 4 (vocals, drums, bass and other). Picking individual instruments takes a paid Premium or Pro plan; it lists 27 stem types in all. An upload counts when it's made, so deleting a song doesn't give the slot back, and the count resets each calendar month. Prices show once you log in. Moises also says it never trains its models on your data. (Sources: moises.ai and help.moises.ai.)
Pick it if you practise every day and want everything in an app on your phone, and five songs a month or a paid plan works for you.
LALAL.AI: the most instrument types
LALAL.AI separates vocals and instruments such as drums, bass, guitar, synths, strings and wind, and adds tools to clean up a voice, take out echo and reverb, and split lead from backing vocals. Files are uploaded to its servers; there are apps for Windows, macOS, Linux, iPhone and Android, and on its Pro plan the desktop app can process on your own computer. Its plans and minutes belong to an account, so we haven't tested it.
Its pricing page, read on 6 November 2026: the free Starter plan gives 10 minutes of processing and files up to 200MB, and its plan table lists result downloads only on the paid plans. Lite is $7.50 a month billed yearly ($90), with 90 fast minutes a month; Pro is $15 a month billed yearly ($180), with 250 fast minutes, the VST plugin and API access. Both paid plans have unlimited minutes in a slower queue. Minutes count as the file's length times the number of stem types you pick. (Source: lalal.ai/pricing.)
Pick it when you need a specific instrument on its own, like the guitar or the strings, and paying for minutes is fine.
Ultimate Vocal Remover: most control, on your computer
Ultimate Vocal Remover (UVR) is a free, open-source desktop app that runs on your own computer, so nothing is uploaded. It offers many separation models and settings, including Demucs, and has no length limit beyond what your computer can handle. It takes an install and some learning, and it's much faster with a good graphics card.
Pick it if you split a lot of songs on a computer and want to fine-tune the result.
Demucs: for developers
Demucs is Meta's open-source separation model, run from the command line with Python. It's the model many of the tools above build on, free, and runs locally. It needs Python and some setup, and a graphics card to be quick.
Pick it if you're comfortable in a terminal and want to script it. If you'd rather not install anything, Extractune runs the same HTDemucs model in the browser.
Tips for the cleanest split, whichever tool you use
Use the best source you have: a clean recording separates better than one filmed off a speaker in a noisy room. Trim to the song before splitting, so the model spends its time on the part you need. And listen before you download, since some songs separate more cleanly than others.
No vocal remover removes noise like wind or hiss, or pulls two singers apart. Where a voice is very loud or has heavy reverb, faint traces can remain in the instrumental with any of them.
Which one should you use?
- 1You don't want your song uploaded, or you're starting from a video: Extractune.
- 2You want a fast split and don't mind uploading: vocalremover.org.
- 3You practice daily and want phone apps: Moises.
- 4You need one specific instrument, like piano or guitar: LALAL.AI.
- 5You split songs in bulk on a computer: Ultimate Vocal Remover, or Demucs if you script.