Real-time · Chrome extension

Hear the dialogue.
Silence the noise.
Hear it, live.

VoiceFront is a free Chrome extension that separates voice from background music and crowd noise as the video plays — not after an upload, not after a wait. Drag a slider, hear the change instantly.

Install from Chrome Web Store — free Who it's for ↓

tabCapture · zero uploads · 100% local processing · MIT licensed

live while you watch

Who it's for

Built for the moment you rewind a third time

😂Stand-up & sitcoms — crowd laughter drowning the punchline? Duck the background, keep the joke.

🎬Heavy background scores — Bollywood, action films, prestige dramas, when the score outweighs the actors.

👂Hearing difficulty — works as a voice lift and a noise muffler at once, no subtitles required.

🌍Watching in a second language — clearer dialogue means less missed meaning.

🌙Low-volume, late-night viewing — keep dialogue intelligible without waking the house.

🎤Karaoke mode — drop Voice to 0% and what's left is a clean instrumental.

The controls

Three sliders. That's the whole interface.

🎙Voice — boosts the isolated centre-channel speech, independent of everything else playing.

🎵Background — controls everything panned wide — score, ambience, crowd. Drop to zero for voice-only.

🎯Sharpness — higher isolates more tightly, with more artefacts. Tune per track for the cleanest split.

See it in action

30 seconds, real video, dragged live

Watch the Background slider drop and the noise fade in real time — no edit, no cut.

Demo · with audio🔊

Video not loading inline? Open it directly on GitHub →

How it works

Position, not magic — and it never stops to think.

Professionally mixed films and songs place the lead voice dead-centre in the stereo field, while music spreads wide. VoiceFront reads that difference live, every 512 samples, with no machine learning and no round-trip to a server.

01 · 02

Capture, then read the stereo field

Chrome's tab capture API streams live audio — nothing leaves your machine. A live FFT checks left vs. right per frequency bin, every 512-sample hop: equal in both channels means centre, meaning voice. Different means panned wide, meaning music or noise.

03 · 04

Mix it back, instantly, your way

Voice and Background become two independent, gain-controlled streams. Move a slider and the change is audible immediately — there's no processing queue, no "rendering," because nothing was ever uploaded in the first place.

No uploads

Audio is processed entirely inside your browser tab. Nothing is sent anywhere.

No AI models

Pure signal processing — a live FFT and a stereo panning mask. No cloud inference.

No server

There's nothing to be down. It runs, fully, on your machine, in real time.

No cost

Free and MIT licensed. Read the source, fork it, build on it.

Get it running