You upload a file you already own, the model splits the vocal off, and in 2 to 3 minutes the instrumental plays in your browser. Nothing to install, nothing to configure, no graphics card.
The separation is done by UVR-MDX-NET-Inst_HQ_3 — the same model file Ultimate Vocal Remover ships with. Not an imitation, not something "similar": that one, running on a server instead of on your machine. The difference isn't the quality of the separation; it's who installs and configures it, and what comes out the other end.
This is what separates a vocal remover from a karaoke generator, and it's worth reading before you make an account:
| What you want to do | Can you? |
|---|---|
| Hear the instrumental on its own, in the browser | yes, there's an "instrumental only" toggle |
| Download the instrumental as an audio file | yes, on Pro — "download the instrumental", in the song menu |
| Download the isolated vocal | no — the separated vocal is never kept: it is used to transcribe the lyrics and deleted afterwards |
| Download a vertical video with the instrumental and the lyrics | yes, recorded as you sing |
| Download the UltraStar .txt score sheet | yes, named "Artist - Title" |
So be straight with yourself: the instrumental does come out as a file here — but only on Pro (US$ 3.99 a month): every song menu has "download the instrumental", and the file arrives named "Artist - Title (instrumental).mp3", the same 192k MP3 the server keeps. On Free the button is there with a Pro badge and takes you to the plans page — the instrumental plays, it just doesn't download. And if your problem is that the file can't leave your machine, no plan fixes that: UVR on your own machine does it for free and nothing is uploaded anywhere.
A separator hands you two files and its job is done. After the separation, this is what keeps happening on its own:
It's the same machine work as always — stripping the vocal — except it doesn't stop at the file.
And the caveat that holds for any separator, ours included: separation never adds quality. If the source file is poor, the instrumental is poor plus separation artefacts.
If your instrumental is coming out muddy and you want to know why before uploading anything, start with why phase inversion sounds muffled. And if your end goal isn't the instrumental but actually singing, the page you want is the karaoke track page.
Yes. You upload the file and the separation runs on the server, with the same model Ultimate Vocal Remover uses (UVR-MDX-NET-Inst_HQ_3). It takes 2 to 3 minutes per song with the cloud accelerator on, and over 10 when it is down.
Yes, on the Pro plan (US$ 3.99 a month). The song menu has “download the instrumental” and the file comes out as “Artist - Title (instrumental).mp3”. On the Free plan the button is shown with a Pro badge and leads to the plans page: there the instrumental plays in the browser and is baked into the vertical video, but never becomes a file. The isolated vocal is not downloadable on any plan — it is not kept on the server.
UVR-MDX-NET-Inst_HQ_3, the same model file distributed with Ultimate Vocal Remover. The separation is the same; what changes is that you install and configure nothing.
It stays in your library on our server, and the heavy part of the processing may run on a GPU service in the United States. If the file can't leave your machine, use a local tool — that's the right answer and we have no way to meet it.
The Free plan does 3 songs a month, gives you 900 MB and asks for no card; the vertical video comes out watermarked. Pro is US$ 3.99 a month, does 30 songs a month, gives 4 GB and the video comes out clean.
StarSinger does those three steps for you. You upload a file you already have, it separates the vocals, transcribes the lyrics with per-syllable timing and gives you a karaoke that plays in the browser, with pitch scoring. You can try it without paying: the Free plan does 3 songs a month and asks for no card — the vertical video comes out watermarked. Past that, Pro is US$ 3.99 a month.
The caveats, because they're real: the transcription gets words wrong on screamed or heavily distorted vocals, and processing takes 2 to 3 minutes per song with the cloud accelerator on, and over 10 when it is down.
Try it free