Setup

Live Answers hears a call through two doors you open yourself, and runs the recogniser in your own browser.

Open a door, then listen

1. Press Share the meeting tab and pick the meeting's tab in Chrome's own dialog, with Share tab audio turned on. Pick the whole screen instead for the system's sound, which is what a desktop meeting app needs.
2. Press Allow the microphone. This second, independent capture separates your voice from everyone else's without guessing.
3. Press Listen.

What a press got is written beside its button — Meeting tab is shared and Microphone allowed — so you can see which doors are open before you start.

A chip then says what is being heard for as long as it is heard: Listening, then the sources — Meeting tab or Entire screen, and microphone — with Pause and Stop listening beside it. A paused page says Paused and transcribes nothing.

While the meeting tab is shared, a refused microphone does not block Listen: the page says only the meeting tab will be heard, and the button is the fix.

The speech model

Speech model offers three, each with the size it downloads once: Standard · 209 MB, Larger · 589 MB, Largest · 1.6 GB. Standard is the default.

The larger two say what they cost. Under Largest: The most capable of the three with names, accents and technical terms, and the slowest: each line takes longer to appear. No accuracy figure is given: none has been measured.

A model this computer cannot run is shown, off, with This computer's graphics hardware cannot run this model. If the page can tell before loading that the one you chose cannot run here, the standard model loads and the page says so; if it fails while loading, press Use the standard model.

The weights download once per browser and are kept: Downloading the recogniser · 209 MB, once, then Loading the recogniser after that.

Requires: a browser with WebGPU. The choice is stored per browser, not per member: what a machine can run is a fact about that machine.

What the workspace sets

Settings → Live Answers is an admin's screen; everybody else reads the values and who can change them:

• Live Answers — on or off for the workspace.
• Recogniser — On this device. Speech is transcribed in the member's browser.
• History — how many days a call's asks are kept. A change applies to calls ending from now on.
• Members may keep a call for good — turning it off refuses new keeps and leaves calls already kept as they are. Keep stays on the row, off, with the reason under the list.
• Company and product names — the names a customer uses for your company and what you sell. Until it is set, the workspace name is used.
• Look up documentation during a call, Docs answers are reused for and Sites to look on — see documentation sources.
• Detect my own asks — for rehearsal: your own questions become cards.

Press Save Live Answers settings to apply them. Under Your calls is one switch of your own, Keep every call of mine, applied the moment it is pressed.