Split any song. Sing it tonight.

From a song file to a karaoke night in four steps

Vocal Split is one app for the whole journey. Every step keeps its result, so you never do the slow part twice.

Separate vocals and music with studio-grade AI

The Splitter page is where every song starts. It uses Demucs, the open AI model that musicians and researchers use for source separation, running entirely on your computer and sped up with Intel OpenVINO.

Choosing and auditioning the song

• Choose file Opens MP3, WAV, FLAC, M4A and OGG files. You can also drag a file onto the app or use Windows' “Open with”. The card shows the name, format, size and length.
• Waveform The whole song is drawn as a waveform. Click anywhere to jump there; drag across it to mark just the part you want to separate. The chosen stretch is highlighted and the times are shown underneath (for example “1:20 – 1:50 selected”).
• ▶ / ❚❚ ■ Every player in the app has the same pair: the round button plays and pauses (pausing keeps your place), the square button stops and goes back to the start of the selection.
• Set start Set end Reset Mark by ear instead of by eye: press Set start and Set end while the song plays to drop the edges exactly where you hear them; Reset goes back to the whole track.

The Output panel: VU meters and a 15-band analyser

Below the player sits an Output panel modelled on classic studio hardware: a left and right VU needle, each with a green lamp that lights while the needle is in the normal range and a red one that lights when it reaches the red, and, between them, a 15-column spectrum analyser with falling peaks. It shows what actually leaves the app — after the equalizer and effects — and it tells you what is playing (“Output · Vocals”).

• + − above and below every column move that equalizer band by 1.5 dB. Hold Shift to push it all the way; right-click a column to set it flat. For a moment after each press the analyser shows the whole curve you have set, so you see the shape rather than one band.

Four ways to separate

Processing

• Start Processing Starts the separation of the marked part (or the whole song). While it works the button fills up like a progress bar.
• Processing Engine A ring with the percentage, the estimated time left, what is being done right now (“Separating audio into stems”, “Clearing the bleed and the noise left in the stems”) and a Cancel button that stops at once and leaves nothing half-written behind.

Results

Every stem arrives as its own card in its own colour — purple for vocals, green for the instrumental, rose for drums, amber for bass, sky blue for the rest and orange for the trimmed original — so you can tell them apart at a glance. The header says how long the separation took.

• Player Play, pause and stop each stem, with its own waveform and time.
• Save Writes the stem where you choose, with a sensible name like “Song (Vocals).mp3” — through your equalizer and effects if you want (see Settings).
• ⋮ Save as… pick another name or folder · Show in folder opens Explorer on the file · Load as source opens that stem on the Splitter so you can separate it again.

The engine panel

• AI Model High Quality (Recommended) — the default Demucs model, included with the app and working without internet. Best Quality (slower) — a fine-tuned model with roughly four times the work for a small gain. Fast (quantised) — quicker and lighter, with a little more bleed. The last two download once, the first time you choose them.
• Acceleration The status card tells you what the separation runs on. On Intel and AMD processors Vocal Split uses OpenVINO, which in our measurements separates about 1.4–1.5× faster than the standard engine with identical results.
• Output Format MP3 320 kbps or 192 kbps, WAV 16-bit or 24-bit, or lossless FLAC.

A cleaner voice and a cleaner backing track

Every AI separation leaves a little of the music in the vocal and a little of the voice in the music. Vocal Split adds its own cleanup pass after the model: it compares the two stems moment by moment and takes out the bleed and the hiss, while leaving everything the model was sure about alone.

• Light (recommended) Removes the bleed and the fizz from both stems without touching what matters.
• Strong Judges each moment harder: cleaner silences, at the risk of thinning a quiet breath or a reverb tail.
• Off Writes the stems exactly as the model produced them.

A 15-band equalizer with a live response curve

Fifteen bands from 25 Hz to 16 kHz, two thirds of an octave apart, each from −12 to +12 dB. Mud at 250 Hz, boxiness at 400 Hz, presence at 2.5 kHz and air at 10 kHz each get a fader of their own. The equalizer applies to everything the app plays — and, if you like, to everything it saves.

• Frequency response The curve on top is drawn from the very same maths as the filter, so what you see is exactly what you hear.
• Faders Drag a fader, turn the mouse wheel over it, double-click to return it to zero, and hold Shift for tenths of a decibel.
• Equalizer on/off switch — compare with and without in one click.
• Preset Nine ready curves: Flat, Vocal lift, Warm, Bright, Air, De-mud, Bass boost, Loudness, Podcast — plus any you save yourself.
• Save preset Delete Reset all bands Keep a curve under your own name, remove it, or start again from flat. Your presets and wherever the faders were left come back the next time you open the app.

Bass you can hear on a laptop, space, width, fades — and any song in the singer's key

The Effects rack runs after the equalizer, in the order shown by the lit signal path at the top: Equalizer → Bass → Gain → Width → Reverb → Fade. Stages that are doing nothing show “bypass”, so you always know what is active.

Bass on small speakers

A laptop, a phone or a television's own speakers play almost nothing below 150–200 Hz — and that is where nearly all of a bass line lives. Separate the bass on a laptop and it can seem to be silent, even though the meters show it moving. Bass on small speakers fixes what you hear rather than what is recorded:

• Harmonics the speaker can play The bass below 130 Hz is turned into its harmonics between 150 and 600 Hz. The ear fills in the missing low note from them — the same trick phones and TVs use — so the bass line is heard on speakers that cannot reproduce it.
• From off to 100% The slider sets how strongly the harmonics come through. They follow the bass up and down rather than growing faster than it, so they do not turn into distortion, and a loud master is softly limited instead of clipping.
• The real bass stays Nothing is taken away: on headphones or proper speakers you still hear the original bass, with the harmonics on top. The effect is timed to the sample, so a saved stem still lines up with the others.
• Small speakers preset — the bass at 70% and the stereo image opened slightly, for listening on a laptop.

Key and tempo

Move anything you listen to up to six semitones up or down and play it at 80–120% of its speed, with a high-quality time-stretch that keeps voices natural. The Original button puts both back. This changes listening only; the karaoke page has its own key and tempo for the video (see below). A track takes a moment to move the first time, then every key you have tried is kept ready.

Turn any separation into a karaoke video

The Karaoke page puts the words over the instrumental as a video for your television — every word lighting up as it is sung, with a countdown before each verse, the next line waiting underneath and a progress bar along the bottom. The result is a standard MP4 (H.264) that plays from a USB stick on practically any TV.

The song and its words

• Which separation Choose any separation from History that has both a vocal and an instrumental.
• Song Artist Guessed from the file name — track numbers, underscores and website tags are cleaned away — and shown on the title card at the start of the video. Correct them if the guess is wrong.
• The words Paste the lyrics, one line per line, exactly as they should appear on screen. The counter shows how many lines there are. Import… reads a .txt file, or an .lrc file which already carries its timing.
• Duets Put who sings on a line of its own before their lines — [Anna], [Mark], or [Both] for lines sung together. Headings like Anna: and [Anna & Mark] are understood too, while [Chorus] or [x2] are simply ignored. Each singer gets a colour of their own in the video.
• Latin letters under the words Off, Automatic, Bulgarian, Serbian or Russian. Adds a smaller line in Latin letters under every Cyrillic line — perfect for friends and family abroad who speak the language but don't read the alphabet. It follows the official transliteration of each language (Щастие → Shtastie, Ђурђевдан → Đurđevdan) and lights up word by word together with the main line. Automatic recognises the language from the letters used.
• Missing lines If some lines in the box are never sung on the recording (a chorus written once too often, for example), Vocal Split tells you and leaves them out rather than showing them over the music. Your text is never changed.

Words that land exactly when they are sung

• Match the words to the voice Uses a forced aligner — an AI that listens to the vocal and finds every word. It rarely needs correcting, and a verse typed once but sung five times is recognised everywhere it comes back, so the box only needs the words, not every repeat. It is a little slower, and needs a 1.2 GB model the first time, downloaded once and kept. With it off, lines are placed on the singing by fast arithmetic.
• Stop after the timing, to check it first On: the big button finds the timing and stops, so you can check it before writing the video. Off: timing and video in one go. The choice is remembered.
• The timeline The vocal's waveform, coloured by who sings: the first singer purple, the second blue, both together green. Click a line to see it in the preview; drag a divider to move where a line starts; right-click a line to change who sings it (first → second → both). Corrections are kept with the song, so the next video starts from them.
• ▶ ■ Play the instrumental — with the guide vocal, key and tempo you chose — and the preview follows along.
• Preview A real frame of the finished video, in its real look and shape, redrawn instantly as you click or drag.
• Make Karaoke Video Asks where to save and writes the video. The button becomes a progress bar that says which step it is on (“Step 1 of 2: finding the timing”, “Step 2 of 2: writing the video”) and how long is left. Afterwards it offers to open the folder.
• Save as MP3+G… Pro Writes the same song as an .mp3 and a .cdg for karaoke players (see MP3+G).

Size, background, key and guide vocal

The right-hand column of the Karaoke page holds everything about how the video comes out.

Real frames, straight from Vocal Split

Every picture below was written by the app itself. Songs, names and lyrics are invented for this page.

Vertical 9:16, laid out for a phone

The phone format is a different picture, not a smaller one. Vocal Split writes it with its own layout: the words take nearly twice the share of the width, wrap sooner, and sit in the middle of the tall frame — ready for TikTok, Instagram Reels and YouTube Shorts.

A karaoke night on your television

The Party page turns your laptop into a karaoke machine. Build a queue of singers and songs, put the words on the TV, and let Vocal Split move from one singer to the next.

The queue

• Add from History Any separated song whose words are timed. It is drawn live on the TV in your Karaoke page look — with count-in, duet colours and Latin letters.
• Add files… Ready karaoke videos (MP4, MKV, MOV, AVI, WEBM, M4V) and MP3+G pairs from any library — pick either the .mp3 or the .cdg.
• Every row A box for who sings, the song and its kind, the singer's key with − +, ↑ ↓ to reorder, and ✕ to remove. Every kind of song can change key — even ready videos and MP3+G.
• Remove all Clears the queue. The queue is kept when you close the app, so a party set up in the afternoon is still there in the evening.

Now playing

• Up next / Singing now Who sings what, the progress through the song and the key. The next song is prepared in advance — key change included — so it starts the instant you press Start.
• Start: name → Pause → Resume One big button for the whole night.
• Restart From the top · Stop ends the song and puts it back at the front, not sung · Skip moves on · − + changes the key while the song plays, without stopping.
• Sung tonight A running list of who sang what.

The television

• Screen Open the screen Connect the TV as a second screen and the words fill it while the queue stays on your laptop. With one screen, the words cover it and the keyboard runs the party.
• Between songs Wait for Start, or start the next song by itself after 5, 10 or 20 seconds.
• Smooth playback The TV picture runs at a steady 25 frames per second in 1080p, drawn separately from the sound so a busy screen never interrupts the music.

A whole folder of songs, done overnight

The Batch page works through a queue of tracks one after another — separating each, and making the karaoke video and MP3+G for every song that has its words beside it. Start it in the evening and find a finished library in the morning.

• Add tracks… Add a folder… Single files, or a whole folder with all its subfolders. Files the app itself wrote are left out automatically, so a folder that contains the last batch's output is never fed back in.
• Words by name Put Song.txt or Song.lrc in the same folder as Song.mp3. Each row shows whether words were found (“words: Song.txt” or “no words beside it — stems only”).
• Status Every row says Waiting, Working, Done, Skipped, Failed or Stopped, with the current step, a progress bar and what came of it. Waiting rows can be removed with ✕.
• Clear finished Remove all Start Batch / Stop Batch Stopping waits for a safe moment and leaves nothing half-written; Start then offers to try the stopped and failed tracks again. One failed song never stops the rest.
• For the whole batch Its own Extraction (including All four parts), AI model, Output format and Cleanup, kept apart from the Splitter's, plus the Output folder with Choose….
• Make karaoke videos For every song with its words, using the Karaoke page's size, background, count-in, guide vocal and look.
• Make MP3+G files An .mp3 and a .cdg for every song with its words — with or without the videos.
• Match the words to the voice The forced aligner, as on the Karaoke page.
• Name the files “Artist - Song” For a library sorted by artist: “01-artist-song (www.site.com).mp3” becomes “Artist - Song”, and the folder, stems, video and MP3+G all take that name.
• Skip tracks already done Anything already in the output folder is left alone, so a stopped batch — or a folder with a few new songs — picks up where the last one finished.

Vocal Split batch\
  Artist - Song\
    Artist - Song - Vocals.mp3
    Artist - Song - Instrumental.mp3
    Artist - Song (karaoke).mp4
    Artist - Song (karaoke).mp3
    Artist - Song (karaoke).cdg

Every separated song also appears in History, and its timing is kept — so the Karaoke page opens it ready to adjust.

Mix the parts — and your own recordings — back together

The Mixer page puts every part of a separation on its own line, with a level, a place between the speakers, a mute and a solo. Add a track of your own — a take you sang, a backing track from another source, a guide — and Vocal Split lines it up with the song by listening to it: where it starts, its speed and its key. Everything is heard as you set it, and the result is saved as one file.

The song and the transport

• Which separation Any separation in History — vocals and instrumental, a single part, or all four parts. For a four-part separation the lines are Vocals, Drums, Bass and Other; the instrumental is left out, since it is those three added together.
• ▶ / ❚❚ ■ Play, pause and stop the mix. The waveform above the tracks is the mix as it is set — click anywhere on it, or on any track's own waveform, to jump there.
• Live Nothing is mixed ahead of time: a mute, a level or a nudge is heard within a fortieth of a second, without the song stopping, and a change of level is ramped so a mute never clicks. The mix plays through the equalizer and effects, at the song's own key and speed.

Every track

Quick mixes

• Everything No vocals Vocals only No drums No bass Drums and bass No other Which parts play, in one press — offered according to the parts the separation has. Levels and your own tracks are left as they are. Practise the bass line with the bass taken out, sing over the band, or hear just the rhythm section.
• Reset levels Every track back to 0 dB, in the middle, unmuted and not soloed.

Your own tracks, lined up automatically

Add a track… takes MP3, WAV, FLAC, M4A, AAC, OGG, Opus, WMA and AIFF files, at any sample rate — several at once if you like. Each is compared with the song and with every one of its parts, and placed against the one it agrees with most: a sung take against the vocal, a backing track against the instrumental. It works in the background, and the track shows Lining it up with the song… until it is done.

• Where it starts A longer intro, a count-in, a track that begins in the middle of the song — found to the millisecond. In our tests the error was under 4 ms.
• Its speed A version played faster or slower — from about 16% slower to 19% faster — is brought to the song's speed.
• Its key Sped up the way a tape runs, key and all, or made in another key — up to six semitones away, which is how backing tracks are often made for a singer — and it is brought back to the song's key.
• Sure or not The rhythm proposes where it goes, and the notes decide: two unrelated songs at the same tempo line their beats up, but not their notes. A track that does not sound like the song is left at the start, with a note saying so, for you to place by ear.
• What was done Written under the track's name — for example “its first 1.454 s come before the song · sped up 3.1%” — and in full on the Your track card, with the part it was matched against.

The Your track card

• Choose a track Click an added track in the list to bring it to the card.
• −100 ms −10 ms +10 ms +100 ms Move it earlier or later by hand, heard at once — the card says how far it has been moved.
• Line it up again Forgets the hand moves and listens once more.

Saving and keeping

• Save the Mix Everything that is switched on, at its levels and places, as one file — MP3, WAV or FLAC, in the format chosen in Settings, and through the equalizer and effects when Settings says so. A mix that would clip is brought down as a whole rather than distorted. The line under the button says what is in the mix.
• Kept with the song The levels, mutes and added tracks are remembered with the separation, so the mix is as you left it next time. The lined-up copy of an added track is kept in the separation's folder and removed with it.
• From History A separation's ⋮ menu has Open in the mixer.

Every separation kept — and cut to length

The History page keeps your recent separations, so you can play and save their stems again without waiting for the AI a second time.

• Open / Close Shows the stems of a run as cards, with play, stop, waveform, Save and their own ⋮ menu.
• ⋮ Make karaoke video opens the run on the Karaoke page · Open in the mixer Pro · Show in folder · Delete this run.
• Clear all Deletes every kept run from this computer, after asking.
• Output meters The same VU needles and analyser as on the Splitter, with the +/− band buttons.

Cut

• Mark a stretch On the stems mixed back together — drag the waveform, or use Set start Set end while it plays; Reset starts again.
• Keep only the selection / Take out the selection Trim a long intro and outro, or remove a solo and join the two sides.
• Fade at the cut From none to 2 seconds — a gentle fade when a stretch is kept, an equal-power crossfade when one is taken out. The panel says how long the result will be.
• Hear the result Save as new run Listen first, then save. Every stem is cut at exactly the same sample, so vocal and instrumental stay in step for karaoke, and the original run is never touched.

Everything in one place

Your settings, presets and the equalizer and effects you left come back exactly as they were the next time you open Vocal Split. The interface is in English.

Free, Home and Pro Editions

Try the whole app for free on a minute of any song. The Home Edition is the complete karaoke maker for your house; the Pro Edition adds what you need to make karaoke for a living. Pricing will be announced at launch.

Built carefully, measured honestly