LUMIÈRE
Lumière · Tools
Who said what. Labelled.
Three voices, two hours, and a client asking who said that. Every segment carries its speaker — detected on your machine, consistent across the whole recording. The audio never leaves it for this step.
How it works
Three steps. No timeline required.
- 01
Bring the recording
Interviews, panels, rushes — however many voices were in the room, long or short.
- 02
Speakers are told apart on-device
Local diarization labels every segment by who is speaking. For this step, the audio stays on your machine.
- 03
The labels ride everywhere
Transcripts, subtitles, text-based trimming — and when you dub, each speaker keeps a voice of their own.
What you get
Built in, not bolted on.
-
On-device diarization
The telling-apart happens locally. Your audio does not travel for it.
-
Whole-recording consistency
The same voice keeps the same label from the first minute to the last — not per-clip guesses.
-
Trim by the right voice
Text-based editing knows who said the removable line — so you cut the words, not the wrong speaker.
-
A cut that obeys names
Because every line knows its speaker, your instructions can too — keep the exchanges between host and guest, leave the moderator out of it. Strike one voice’s lines in the text editor, or tell the story brief who carries the film.
-
Per-speaker dubbing voices
In a dubbed master, each speaker gets their own voice instead of one narrator flattening the room.
Speakers are numbered labels today — Speaker 1, Speaker 2. Naming them once, and having the suite remember people, animals and objects everywhere, is in development — not for sale yet.
One suite, every tool
The same engine, other doors.
Each tool here is the same production suite, entered where you need it. Start with one; the rest is already there.