Choose how a recording shows

Agents: claude, codex, opencode · On: phone, desktop · optional

The VCode machine turns every dictation into text. The cog's Show the recording as picks what you watch while you speak: a waveform, or the words typed as you say them.

Why you would use it

The waveform records a clip and shows its level; the words arrive when you stop. Live text shows the words as you speak, so you can stop as soon as they read right; when you stop, the machine replaces them with its own punctuated text. Live text downloads an 86 MB speech model to the phone once, and is offered only on a box that has it — see dictate-live.

How to use it

Tap the cog (#settings-open).
In the Voice group (#settings-voice), find the Show the recording as row. Its control (#voice-display) is two segments: Waveform and Live text.
Tap Waveform or Live text.
The choice saves at once and applies to the next recording.

On a box without the live engine there is one segment, Waveform, and it is disabled.

What you see

With Waveform, a tap on the mic records a clip: a strip of bars takes the place of the composer's text line, or sits under its text if it has some. The strip is 48 px tall on a phone; over an empty composer on a wide screen it is the height of the composer's one line. See dictate-a-message.

With Live text, the composer's text stays in view and the words appear in it as you speak, with no waveform. The first tap after a fresh browser downloads the model: the mic shows a download arrow with a filling ring and Downloading the speech model, 42%. See dictate-live.

The right end of the Voice title (#voice-hint) shows Saving…, then a green check and Saved. On a box with the transcription service set up, it reads Sidecar set up with a green check whenever it has nothing newer to say. That means VCODE_STT_URL is set, not that the sidecar answered.

Options and settings

Option	Default	What it changes
`#voice-display` (`voice.display`)	none saved, which is `waveform`	`waveform` or `live`. Anything else is refused with `voice.display must be one of waveform, live`
`live` saved on a box or a browser without the live engine	—	Records a waveform clip instead, rather than leaving a mic that cannot start

A settings.json written by a build that had the Engine pick still loads: engine: "live" becomes display: "live", and the other engines and the old correction switch are dropped. A save that still names engine or liveCorrect is refused with unknown voice setting: engine.

Limits and known gaps

The choice lives in settings.json on the box, so it is shared by every device signed into this box. A phone cannot hold its own choice.
Live text needs Worker, AudioWorkletNode and WebAssembly in the browser. Without them the row offers only Waveform.
There is no mic at all without the transcription service, live engine or not: the machine transcribes every dictation. See install-the-voice-sidecar.
The app on app.v-code.dev offers only Waveform — see use-the-app.
With Waveform, a tap while the app is offline is refused with offline — dictation needs the server. Live text still starts offline once its model is loaded in the page, and skips the correction: the live words stay as they are.