Voice input for AI Elements
The point of this experiment: replace the browser's built-in Web Speech mic with real API transcription. That trade buys cross-browser behaviour (no Chrome-only SpeechRecognition), an actual model choice, a latency/accuracy delay dial, and pre-roll — the mic captures from the instant it is pressed, so nothing said during connection setup is lost.
gpt-live-transcribe · useVoiceInput · ws-preroll · AI Elements PromptInput
Prompt input
While listening, completed utterances are committed into the box; the tail you are still speaking is appended live, and typing is paused — mixing manual edits with a live machine-appended tail would put words in unpredictable places, so stop the mic to edit. A controlled textarea cannot style part of its value, so interim text looks like the rest — the amber indicator is what tells you the tail is still provisional.
Submitted
There is no chat backend here — the experiment is the input. Submitting just echoes the message below.
nothing submitted yet