-
Microphone on stage
Someone speaks. Your sound team already has the signal.
-
Sound console
A clean speech feed goes out: a post-fader aux, not the room or the music mix.
-
LCS Box or your computer
Speech becomes text, and text is translated, on the machine beside you.
-
Every output
The same caption, at the same moment.
- Welcome, everyone.Projector screen
- Key/FillWelcome, everyone.Video switcher Key/Fill ATEM
- CCES · Bienvenidos a todos.Stream YouTube captions, OBS, vMix
- FR · Bienvenue à toutes et à tous.Phones Scan the QR code
-
One speech feed, four stages
The planned path starts at the microphone, before the room loudspeakers. Route only the people speaking to a dedicated post-fader aux or matrix. Send that bus to the caption computer; do not use a room microphone or the full music mix.
- Audio in → the dedicated speech bus reaches the computer through console USB audio or an audio interface.
- Speech model → speech recognition runs on the computer’s GPU and produces the original caption text.
- Translation → finished sentences are translated locally, with French ↔ English planned first.
- Outputs → the original and translated text go to the selected screens, stream overlays and phone viewer.
Offline at show time
Live Caption Studio is built to run with no internet at show time. Free needs no activation and works offline indefinitely. One-time licenses work offline indefinitely on the purchased major version after a single activation (online, or by file from another device). Subscriptions renew once a month, online or by license file, with a 7-day grace; without renewal, the installation then continues as Free. There is no cloud speech recognition or translation fallback in the design.
The only network calls are license activation/renewal and update checks, and both can be turned off. YouTube captions, if you turn them on, send caption text (never audio) to YouTube. Local processing does not make an internet delivery destination work offline.
Where captions are planned to go
For audience screens, the plan includes a ProPresenter overlay and a phone or tablet viewer reached by QR code. OBS and vMix use an overlay page as a browser source. YouTube Live closed captions are a separate text output.
For an ATEM switcher, the planned linear key path uses a fill image and a separate key image over two HDMI outputs. Those outputs need to be available on the caption computer. These integrations are planned for version 1.0; they are not a list of shipped features.
What determines delay?
The speech model needs audio before it can settle words. Translation then works on finished sentences; rendering and the output path add their own time. Caption delay and translated-line delay are different measurements.
We will publish measured delay and accuracy with the hardware, date and method before quoting a product number. No published product-level measurement is claimed here. Synthetic captions on this website illustrate the layout; they are not a performance demonstration.
Plan the feed and the computer together
Live Caption Studio needs a clean speech feed: the microphone routed directly from your sound console (a post-fader aux or matrix), not the room or the music mix. We make no accuracy promise for speech over a music bed.
The planned computer requirement is NVIDIA RTX on Windows or Linux, or Apple Silicon on a Mac. Check the tier for captions alone versus captions with translation before proposing a rig for the beta.
For best results, keep speech loud enough, the microphone close and steady, and one voice speaking at a time; overlapping speakers reduce recognition accuracy.