Free tool
Vocal remover
Upload a song and hear it split into the instrumental and the vocal — the same separation the Studio runs, played back against the original at matched loudness so you judge the split and not the volume. Hearing it is free. Keeping the files is paid, per track, and the price is on the button.
Uploaded, processed, auto-deleted within 24 hours. Unlike our other tools, this one sends the file to our servers, because the separation runs on the engine. The upload and both parts are deleted automatically within twenty-four hours — unless you sign up and move the result into your account, where it is kept like any other file in a workspace. See the privacy policy. You must hold the rights to any audio you upload; the acceptable use policy applies, and rights holders can reach us at /legal/copyright.
Drop a song, or choose one. It is uploaded, separated on our servers, and deleted within 24 hours. Up to 15 minutes.
If two parts are not enough
The same engine splits into four, five or six
This tool does one two-stem split per song. The Studio does the rest.
Drums, bass, piano, guitar and the rest — the Studio’s stem separation runs the same engine over the same upload and hands back every part at full rate, in the priority queue, with the project kept so a master or a mix can start from the stems. The free plan includes it; every capability is available on every plan.
The method
How a vocal is removed from a finished mix
What the engine is actually doing, what it can and can't recover, and how to read what you hear.
A mixed song is one waveform in which every instrument overlaps in time and in frequency, so there's no vocal “track” left inside it to pull out. A vocal remover instead estimates, moment by moment, which part of the sound is the voice, writes that estimate out as the acapella, and subtracts it from the mix to leave the instrumental. The quality of the split is the quality of that estimate.
- Why phase cancellation isn't enough
- The old trick — flip one channel and sum, so anything panned dead-centre cancels — takes out the lead vocal only if it was centred, and takes the kick, bass and snare with it, because they are centred too. It also leaves every reverb tail and every double-tracked harmony in place. It's fast, free, and the reason people expect vocal removers to sound hollow.
- What a learned separator does instead
- The Euphona engine runs a model trained to recognise what a voice looks like in a spectrogram — its harmonics, its vibrato, the way consonants sit above the band — and to mask everything else out. It follows the vocal wherever it is panned, keeps the drums where they were, and hands back two parts that add back up to the original, which is the property that makes an A/B honest.
- What to listen for
- Switch between the original and the instrumental on a sustained note and on a breath. A good split leaves no ghost of the melody in the instrumental and no hi-hat leaking into the vocal. Reverb is the hardest case: the voice’s own reflections belong to the voice, and a separator has to decide where the room stops being the singer.
- Where it will still fall short
- Heavily distorted vocals, a lead doubled by a synth playing the same line, or a mix already crushed to a lossy stream give the model less to go on, and the artefacts show up as a faint watery texture on the quietest passages. Start from the best copy of the song you have — a WAV or FLAC over an MP3 — and the split improves with it.
The longer explanation covers how multi-stem separation differs from a two-way split, how the parts are measured, and what “the stems sum back to the source” means in practice.
Read about stem separationQuestions
What people ask before uploading
- Is it free?
- Hearing the result is free, with no account, no email and no watermark. The full-resolution files are paid — priced per track from the length of the song, between $2 and $8, or included in a subscription. Signing up moves this result into your new account, where the download button quotes the same price. There is no bait-and-switch at the button because the price is on it.
- What happens to my file?
- It is uploaded to our servers, separated, and deleted automatically within twenty-four hours together with both parts. Signing up moves the result into your account, and from then on it is kept like any other file in a workspace. Nobody on our staff can listen to it. It is never used to train anything.
- How long does it take?
- Free jobs run in the gaps between paid work rather than ahead of it, so the wait depends on how busy the engine is — usually a couple of minutes for a four-minute song, sometimes longer. We publish no time promise for this tool because we have not measured one yet.
- Why is there a 15-minute limit?
- Separation is the most expensive thing the engine does per minute of audio, and the tool runs on a monthly budget of engine time. Fifteen minutes covers almost every song; longer material runs in the Studio, where the limit is an hour.
- Is the result made by AI?
- Yes. The split is estimated by a machine-learning model, and both files carry a machine-readable mark saying so, applied when they are written and checked after every conversion. It is a tag, not a watermark — nothing is added to the audio. The notice is at /legal/ai-transparency.
- Is this a vocal isolator as well?
- Yes. One split produces both parts, so the same upload answers both questions: remove the lead vocals from a song and keep the instrumental, or isolate the voice and keep the acapella. A tool calling itself a voice isolator or an acapella extractor is doing this same two-stem job, read from the other side.
- Can I upload any song?
- Only audio you hold the rights to process. Most people use this on their own recordings, on tracks they are remixing with permission, or on material for practice and study; the acceptable use policy says where the line is, and the twenty-four-hour deletion is how we keep the tool from becoming a library of other people’s work.