Vocal Reduction & Lyric Timing
Reduce centre vocals from a stereo file in your browser. Export WAV and time lyric lines manually. No AI stem separation or uploads.
Open the workbenchIn-browser centre-channel reducer
Karaoke workbench
Reduce the centre of a stereo mix, export the result as WAV, and time your own lyric lines by tapping along. Decoding, processing and encoding all happen in this browser tab.
- Runs on your device — nothing is uploaded
- Not AI stem separation
- Stereo only
Step 1 · Source track
Drop a stereo audio file here
or click to choose a file
MP3, WAV, M4A, OGG, FLAC · decoded locally, never uploaded · Up to 64 MB and 8 minutes
Step 2 · Centre reduction
Reduction strength
-10.5 dB on the centre band
Preserve low end
Bass and kick usually sit dead centre too, so a full-band cut hollows out the bottom. This leaves everything below the crossover centred.
Normalise output to −1 dBFS
Scales the rendered track to a fixed peak so the export is not quieter than it needs to be.
What this is — and what it is not
- What it does
- It attenuates the centre (L/R-correlated) part of a stereo mix — the classic mid/side vocal-cancel. A lead vocal panned centre usually drops a lot; the sides stay.
- What it is not
- Not AI stem separation. There is no trained model here: it cannot isolate a vocal, and it cannot remove one that is not centred.
- Stereo only
- A mono file has no side channel, so there is nothing to split. Export a stereo version of the recording first.
- Quality depends on the mix
- A dry, centred lead over a wide backing works best. Hard-panned, doubled or reverb-heavy vocals only partly reduce, and the result can sound hollow or phasey with a thinner low end.
- Nothing leaves this device
- The file you pick is decoded, processed and encoded in this tab. It is never uploaded, and no audio is fetched from anywhere.
- Your rights, your call
- Use it on recordings you are allowed to process — your own, or ones you have permission for. This page grants no rights in anyone else’s song.
Step 3 · Preview and export
Run the reduction to preview and export a karaoke mix.
No track loaded yet.
Step 4 · Lyric timing
Mark each line while it plays
Paste your own lyrics, play the track, and stamp every line as it starts. Timings come only from your taps — there is no automatic lyric alignment here.
Load a track above to play along while marking lines.
A signal-processing tool, not a karaoke catalogue. A licensed karaoke release of a commercial song is still the only version that comes with permission built in.
How to use this tool
"AI karaoke maker" usually means a model that pulls the lead vocal out of a finished song. There is a real tool on this page, and it does something narrower than that phrase promises: it reduces the centre channel of a stereo file you own, entirely in your browser, and lets you time the lyrics yourself. No upload, no machine-learning stem separation — and no right to publish somebody else's recording just because the vocal got quieter.
What the karaoke workbench does
Everything runs on your device. You pick a local file — MP3, WAV, M4A, OGG, or FLAC, up to 64 MB and 8 minutes long — and the browser decodes it with the Web Audio API. Nothing is uploaded, and using the tool does not spend credits.
The processing is centre-channel reduction. A stereo mix puts whatever both channels share in the middle: the lead vocal usually sits there, while guitars, keys, and wide drums spread out to the sides. The tool subtracts that middle signal (a mid/side filter), which lowers what is centred and leaves the sides closer to intact. The controls are:
- Reduction strength, 0-100%, with Light, Balanced, and Strong presets — how far the centre is pushed down.
- Preserve low end, with a crossover frequency — so the kick and bass, which are also usually centred, are not thinned out along with the voice.
- Normalise output to -1 dBFS, if you want a consistent, near-peak level.
Then you download a WAV — 16-bit PCM or 32-bit float, named after the file you loaded. That is the whole shape of the tool.
Say the important part plainly: this is a fixed mid/side filter, not a model. It does not recognise a voice, it does not split a song into drums, bass, and melody, and it cannot remove a vocal that was not mixed into the centre.
Stereo only, and what centre reduction can and cannot do
Two rules are stated in the tool: a mono file cannot be processed, because there is no centre channel to reduce, and a file with more than two channels is read from channels 1-2.
Inside those rules, results depend on the mix:
- Good candidates: a stereo master with a dry lead vocal mixed dead centre and instruments spread wide. Balanced strength is usually the starting point, and it is worth comparing Light and Strong before you export.
- Hard cases: heavy reverb or delay on the vocal, doubled and widened vocals, a voice sitting slightly off-centre, mono masters, and anything where the lead itself was spread across the stereo field.
- What always comes along: bass, kick, snare, and any solo instrument panned centre get pushed down with the voice. Preserve low end keeps the bottom octave; it cannot separate a centred guitar from a centred voice.
Pulling a vocal out of one stereo file is always an approximation. If you still have the multitrack, muting the vocal track in a DAW will beat this every time. If the recording is not yours, read the licensing note below before you do anything with the result.
Timed lyrics: you mark the lines yourself
The workbench also ships a lyric timing editor, and it is deliberately manual. Paste your lyrics, one line each, play the track, and mark each line as it starts — click Mark line N at 0:12.34 or press T. There is no automatic alignment and no transcription, which is why the timings follow what you actually hear instead of a guess about it.
When the pass is done, export a .lrc file or copy the timings as text. Most players read .lrc, so the file works in any karaoke app that supports synced lyrics, and you can keep it next to the WAV you exported. Use lyrics you have the right to use: writing down the words of a song you did not write does not change who owns them.
Karaoke of a commercial song is still a licensing question
Lowering the vocal in a copy of a recording does not lower the rights attached to it. The composition and the recording belong to their owners, and a centre-channel filter changes the file, not that fact.
- Releasing, distributing, or performing it publicly needs permission from the rights holders. A licensed karaoke service or an authorized karaoke release is the route that comes with permission included.
- Practising and singing along privately is where this tool is comfortable.
- The acceptable use policy states where OnePromptSong draws the line on copyright and impersonation, and the terms of service cover the product itself. Neither grants rights in someone else's song.
Making an original song here instead
If what you want is a song you can keep, publish, and send, the studio builds one from your own words: describe the song on the home page or open the studio, confirm the brief, rewrite the lyric draft, then generate two original versions with vocals — delivered as a full-length MP3 and a private share page you can switch off at any time. One version costs 10 credits, a fresh A/B pair costs 20, and keeping or sharing a finished song costs nothing extra; the pricing page lists the current plans. That song is generated for you, so there is no third party to clear it with, and the usual caveats about AI output still apply.
Two craft notes if you plan to sing it yourself:
- Short lines and a repeated hook are what make a song singable by someone who has heard it twice. Long verses are for the performer; the chorus is for everyone else.
- Write the words the way you say them. A line you plan to sing has to fit one breath, so read it at tempo before you generate and cut whatever makes you stumble.
Start from the song maker home if you want the overview first. Related reading: rap lyrics generator, instrumental maker, Suno alternatives, credits and pricing.
