Make a karaoke track, clean up your voice, write to your beat

Yes. No paid tier, no watermark. Without an account you get up to 3 tracks a day; a free sign-in with Telegram or Google gives you up to 40 an hour. The limits keep the server fast; they are not there to sell you anything.

It is written to a temporary folder on the server, processed, and deleted two hours later along with the results. It is never sent to another company, and the download links are unguessable random tokens.

Split the song, then pull the backing fader all the way down: what is left is the vocal on its own. Download the vocal stem as WAV or MP3 for a remix, a cover or a DJ edit. It keeps the singer's reverb, and on a busy mix faint traces of the instruments can stay in it.

Yes, that is what the two faders under the player are for. They change the balance while the song plays, and the button next to them saves what you hear as one file: a karaoke track with a faint guide vocal, or a loud vocal pulled down a few dB.

Of almost any. The cleanest results come from studio recordings with one lead vocal. Dense choruses with stacked harmonies, heavy reverb or echo on the voice, and live recordings are harder: faint traces of the vocal can stay in the backing. It is free, so the quickest test is the song you have in mind.

Yes. After the split, “Sing it in your key” under the player reads the singer's range from the vocal and compares it with yours (the warm-up measures it in about a minute and keeps it in your browser). It suggests a shift of up to 6 semitones either way and renders the instrumental in that key at the same tempo, to listen to and download. Each render counts as one track against your limit, and past three semitones the backing starts to sound artificial.

Yes, two ways. If you have the song with vocals, split it here: once the warm-up has measured your range, the site suggests the shift and renders the instrumental in your key. If you already have a karaoke track, the pitch & tempo changer moves it to your key in semitone steps, up to an octave either way, without changing its speed.

A semitone or two is hard to hear. Past three or four the instruments change colour and start to sound artificial, so a backing moved that far is better for practice than for a performance. The site suggests at most 6 semitones either way; if your voice needs more, singing an octave lower or higher usually sounds better.

For most produced music the instrumental is clean enough to sing over and the vocal is usable on its own. Dense choruses, heavy reverb and live recordings are the hard cases and can leave faint traces.

MP3, WAV, M4A, FLAC, OGG and most video files: the soundtrack is pulled out automatically. The limit is 60 MB and 6 minutes.

About one and a half times the song's length: the neural network runs on an ordinary processor, not a graphics card, so a four-minute song takes about six minutes. The page shows the progress and your place in line.