Voice input in the web composer, using the browser's speech recognition

[Dwite]

The problem

Voice input works on iPhone only. In the web app I can only type.

What I would like

A microphone button in the web composer, in browsers that support speech recognition.

• Press it to start. Then finish or cancel.
• The text goes into the composer so I can review it. Nothing is sent by itself.
• A browser without support shows a clear "not supported" state.

How it would work

It uses the browser's built-in SpeechRecognition API. It needs no API key, no new server part, and no provider change.

Things you should know before you decide

• Privacy: the browser may send the audio to its own speech service. On iPhone the audio stays on the device, so this is a real difference. The button, the setting, and the user guide would have to say it clearly.
• It needs HTTPS or localhost.
• The desktop app has no speech service, so it would not get this feature. It would show as not supported.
• This does not add push-to-talk or more languages.

What I am asking

• Is browser speech recognition acceptable for T3 Code, given that audio can leave the device?
• If yes, is "web browsers only, always review before sending" the right first scope?

I built a version before asking, which was my mistake. It is closed, but it has demo videos: #12893. That PR also had an iPhone "send immediately" setting. I moved that part to its own small PR, #14807, and it is not part of this request.

Related: #10489 (desktop voice and Chinese), #7417 (push-to-talk).