By Sam Ridder — I build and run StemGrab on my own hardware, on my own. Who I am.
Upload a track and get a clean instrumental plus the isolated vocal in one go. Comparing it to the paid apps first? See how this stacks up as a Moises alternative before you commit to a subscription.
Only on Spotify or YouTube? You need the audio file itself — record it with our voice recorder or extract it from a video file.
Real stem separation on a GPU, no sign-up, no watermark, 10 tracks a day.
The fastest way to remove vocals from a song without paying for software or hunting down a stripped-down file: you get a clean instrumental plus the isolated acapella. This song vocal remover (some call it a voice deleter) works on the file you already have, so you are not limited to songs that happen to have an official instrumental.
Full-quality stems. No account, no per-track fee. Same model and same 192 kbps (or lossless WAV) as the paid tier — tracks up to 15 minutes.
A typical three-to-four-minute track takes six to ten seconds, on a phone or a desktop, and on most modern mixes the instrumental is clean enough to sing over.
Your upload is deleted the moment the split is finished. Want WAV, play-along mixes or longer tracks? The full tool is on the home page.
The vocal remover AI model runs on our own GPU and is trained specifically on separating voices. The AI is trained on what a voice actually sounds like, so it lifts the singing out instead of cancelling whatever sits dead-centre in the mix the way the old phase-inversion trick did. That's why one upload returns both a clean instrumental and clean isolated vocals, whether the track is mono or stereo, dry or drenched in reverb. This page uses a dedicated vocal AI model that is cleaner on voices than the general splitter on the home page, so it is the better choice when the voice is all you care about.
Every upload here returns just two files: the vocal and the instrumental. The full splitter on the home page returns drums, bass, guitar, piano and other as separate 192 kbps MP3 files too. Need only the vocal extractor output? Use the acapella extractor. This is an online vocal remover tool: nothing installs on your machine, so you don't need vocal remover software or a vocal remover app; the browser is enough. If you're weighing it against the LALAL AI vocal remover, the LALAL.AI alternative page puts the two side by side.
The vocal remover is the single most-used tool on the site: 1,776 uploads in the 40 days we've been measuring, more than every other tool here combined. Looking at the files themselves — 320 uploads measured by size, sitewide — the average was 9.44 MB and the largest was 83.66 MB; most cluster between 4 and 8 MB (107 of them), 8 to 16 MB (57) and 2 to 4 MB (43), with 98% comfortably inside our limits rather than pushing against them. That range is what a normal three-to-five-minute MP3 looks like at everyday export quality — not a red flag, just what an average upload is. What goes wrong is duller than the separation itself: the most common upload rejection anywhere on the site was a file with no extension at all, 26 times in 40 days — usually a document or a renamed file dropped into the box, not a format we don't support. If your upload gets turned away, that's the far more likely reason than anything wrong with the vocal removal itself.
Big files are fine. The free tier takes up to 2 GB per upload, so a 24-bit studio WAV goes in as easily as a phone recording, and it runs on the same graphics cards as a small MP3. The limit that matters is length: 15 minutes of audio per file. A long DJ set or a full album side hits that long before it gets anywhere near 2 GB.
Most free "vocal remover" tools you'll find online are still built on phase inversion: flip the polarity of one stereo channel, sum it with the other, and anything panned dead-centre cancels out. It's a neat party trick, but it isn't really vocal removal: it cancels the kick drum and bass right along with the singer, it only works on mixes panned exactly down the middle, and it fails outright on mono files or vocals carrying stereo reverb. What's left is a hollow, phasey mix with the vocal only partly gone, not a real instrumental.
StemGrab doesn't cancel anything. Can AI remove vocals better than that? In practice, yes: the model is trained to recognise what a voice actually sounds like (its formants, breath and phrasing) and lifts that voice out of the mix directly, whether the track is mono or stereo, centred or panned, dry or soaked in reverb. That's why one upload gives you a genuinely clean acapella and a genuinely clean instrumental instead of a compromise. Curious about the mechanics behind it? We wrote up how AI stem separation actually works, phase tricks included.
Most people who remove vocals online want one of two things: a karaoke track to sing over, or an acapella to remix. To remove vocals from a song online you need three steps and no account.
That's the whole routine for how to remove vocals from music, and it is the same answer to "how can I remove vocals from a song on my phone?", because the page runs in any mobile browser. For a karaoke night you can take the vocal remover from song to song, up to 10 tracks a day. People who have used vocalremover online elsewhere will recognise the flow; the difference here is a dedicated vocal model, no account and no watermark. Spanish- and Portuguese-speaking visitors often know it as a remover vocal online, and this page exists in both languages (see the language menu). Expect a little more bleed when you remove vocals from music with a very dense mix; on a sparse acoustic track the voice comes out almost spotless. How to remove vocal from music with stacked harmonies is covered under "Where vocal removal still struggles" below.
It's a vocal remover and isolation tool in one, so every upload gives you two files:
The tool works in both directions. Use it to remove voice from song files and you keep the instrumental, which is what a voice remover from music is for. Keep the other file instead and it works as a music remover from audio: the band goes, the singing or speaking voice stays. Want to remove instrumental from song files and keep only the acapella, or rip vocals for a mashup? Download the vocal stem and skip the other one. Used as an instrumental remover free of charge, it gives the same quality as the paid tier. An acapella remover in the literal sense is simply the instrumental file: if you want to remove acapella from a track and keep the band, download that one. Want to isolate vocals for an a cappella cover instead? Keep the vocal file. There is no separate mode to switch on for either.
Planning to publish what you make (a remix, a mashup, a cover over the instrumental)? Read what you can legally do with the result first; practising and experimenting privately is always fine, publishing someone else's recording usually needs permission.
Working from a video instead of a song, say a vlog or tutorial with music mixed under your own voice? Drop the video straight in here and keep the acapella track as your narration. Editing in GarageBand? Separate the song here first, then drag the instrumental straight into your project; no plugin is needed inside the app.
No separation model is perfect, and it's worth knowing which songs will fight back before you blame the upload. Stacked harmony vocals are still vocals as far as the model is concerned, so they leave the mix along with the lead singer: if the chorus you wanted to keep is three voices thick, your instrumental will suddenly feel empty right where the song should open up. Long reverb tails are the second one. The voice is dry when it's sung but the room it was sung in is part of the recording, so on a track drenched in hall reverb you can get a faint ghost of the vocal hanging in the instrumental even though the voice itself is gone.
Vocoders, talk-box and heavily pitched vocals read as instruments rather than voices, which is usually the wrong call for a modern pop hook: expect part of that hook to stay in the instrumental. Live recordings are hardest of all, because the crowd, the room and the bleed between microphones all sit on top of each other. And if you remove vocals from song recordings made before roughly 1970, where the whole band was often cut to a narrow mono mix, there is simply less spatial information for the model to work with.
A harmony vocal that sits tight against the lead, even a single line doubling the melody, almost always leaves the mix together with it. When you remove vocal from song files with a choir or backing singers, all of them go: to the model, any singing is still singing. If you want to remove background vocals from a song without touching the lead line, that's not something stem separation can do reliably yet.
None of this makes the result unusable. It means you should listen before you commit: a dense live track might need the vocal stem pulled down rather than removed, which is exactly what the mixer is for.
StemGrab doesn't hand you a single "vocals removed" file and hope for the best. When the job finishes you get the instrumental and the vocal as two separate files, so you can play the instrumental on its own through the section you actually care about. That is the fastest way to catch the problems above: skip to the chorus of the instrumental and you'll know within ten seconds whether the backing harmonies went with them.
If it sounds right, download just the file you need instead of the zip. If it doesn't, you now know why, and whether a different section of the song works better. Want the vocal melody itself as notes rather than just a key label? MP3 to MIDI turns the same upload into a MIDI file with the vocal line on its own track.
Yes. The ads on the site keep the GPU running, so you can split vocals and instrumental free: up to 10 tracks a day and 15 minutes per track, with no watermark and no trial that runs out. You can use the vocal remover free no sign up required, and the download starts straight from the page. A paid pass adds longer tracks, practice tools and no queue; the separation model is the same.
On most modern, well-mixed tracks the separation is remarkably clean, and far better than the phase-inversion tricks older tools use. Very dense mixes, heavy reverb or a live recording with crowd noise are harder for any vocals remover or vocals separator: expect a faint residue there rather than a perfect split.
On most modern, well-mixed songs, yes: the output is 192 kbps MP3 with no audible phase artifacts, the same model and bitrate as the paid tier. If you want lossless WAV, the full tool on the home page offers it.
No. The model treats all singing as one group, so backing vocals and harmonies leave the mix together with the lead; there is no background vocal remover setting that splits them. If you only want to remove the vocals partly, turn the vocal stem down in the results-page mixer instead of muting it.
Not on this page: as a vocal splitter it returns the vocal and the instrumental. If you want a stem vocal remover that hands back every instrument, the full stem splitter free on the home page returns vocals, drums, bass, guitar, piano, other and the instrumental from a single pass, so use that when you want each instrument as its own file.
MP3, WAV, FLAC, M4A, AAC and OGG, plus video: MP4, MOV and MKV go in as-is, so it works as a video vocal remover without pulling the audio out first. As an MP3 vocal remover it handles already-compressed files fine. Files can be up to 2 GB and 15 minutes long. The download is always audio (the separated stems), not a video file.
No. This isn't a YouTube vocal remover that takes links; this vocal remover website works on the audio or video file itself. If you only have a song on a streaming service, you need a file you are allowed to use first. You can extract the audio from a video file you already have.
Yes. Using the vocal remover on a phone works the same as on a desktop: open the page in any mobile browser, pick the file and download the result. You get the same vocal remover free online on any device, with no app to buy.
No. There is no voice remover software to download and no browser extension; the separation runs on our own server and you use it from the page. You don't need any other software for vocal remover jobs either.
Vocal extraction takes about six to ten seconds for a typical three-to-four-minute track on our GPU. Longer files take a little longer, and at a busy moment you may wait in a short queue first.
Yes. You can remove vocal song after song: when one file finishes, drop the next one in. Each run is independent, and the voice remover free tier allows up to 10 tracks a day.
Your upload is deleted right after processing and the stems are wiped within 45 minutes (24 hours with any pass, paid or free — that longer window is part of what the pass is for). See the privacy policy.
A little, on some songs. The model separates sounds that were mixed together, so on a dense track a faint residue can remain in the low-mid range. On a typical studio mix the difference from the original is barely audible.
Yes. It can separate voice from music in speech too: if a recording mixes a speaking voice with a background track, the tool lifts the two apart so you can keep the voice alone. Used to remove music from a podcast or narration, it is a music remover free of charge for spoken recordings. It also works the other way round: remove voice from mp3 interviews and keep only the music bed. For a video, remove background music from video is built for exactly that.
A pre-made one is an instrumental released by the label. A vocal remover separates the voice from a normal stereo mix after the fact, so it works on any song you already have.
Yes, in two steps. Separate the song here first, then run either file through the pitch changer to change the pitch, or read its tempo in the BPM finder. Any pitch shifter works better on a separated file than on the full mix. To stitch pieces back together afterwards, use the audio joiner.
The tool is free to use, but the rights to the song stay with whoever owns them: for anything public, get permission first. The legal guide explains what that means in practice.
Ultimate Vocal Remover (UVR) is open-source desktop software. A UVR vocal remover install needs a decent graphics card and some setup, and you choose and manage the models yourself. StemGrab does the same kind of separation in a browser tab, with nothing to download. The LALAL AI vocal remover uses a similar kind of AI model but sells processing per minute; here there are no credits and no paid plan needed just to try it.
My Edit vocal remover caps free downloads at one a day and tracks at ten minutes each; this vocal remover online free of charge gives you 10 tracks a day and a 15-minute ceiling per file.
Downloaded software is where the comparison with Yogen starts: you install the vocal remover Yogen offers on Windows or Android before the first file even loads. The Yogen vocal remover also only works well when the lead vocal sits dead-centre in the stereo mix, because it recentres the channel rather than running a trained separation model, so an off-centre vocal often survives untouched.
The Digital Magic Wand vocal remover, a feature the site calls UnMixIt, sits inside a wider suite that also edits images, video and text; this page does one job, splitting a song into vocal and instrumental.
X-Minus caps its free tier at a handful of minutes per track and a rolling daily total; its paid pro tier starts around $2.48 a month for higher limits. This page is a free vocal remover with no account needed at any tier.
ReMusic AI (often searched with a doubled "ai ai" by mistake) runs on its own servers, needs no sign-up and separates a track in about ten seconds, roughly the speed of the run above. Phonicmind lets you preview a separation without an account, but downloading the finished file needs a paid plan; every download on this page, not just the preview, is free.
By Sam Ridder — I build and run StemGrab on my own hardware, on my own. Who I am.
ShinobiTools · KvK 42147182 · info@shinobitools.com