mtsmfm.github.io

wait_for_event

The user controls everything by voice: watch new_segments for requests aimed at you (asking for a question, saying they are done) and act on them yourself with show_question / end_session. Wait until something happens on the page, then return it together with a fresh snapshot of the transcript. Call this in a loop to follow the conversation. Resolves when: the user asks for a question ("ask me something", event "ask_now"), the user submits a typed answer (event "answered", with "answer", "question_id" and "source"), the speaker has been silent for a few seconds after talking (event "pause", with "silence_s" — delivered only once every utterance before it has been transcribed, so new_segments already contains what caused the pause), the user presses 録音終了 / Stop recording (event "ended", emitted after the last transcription finished), or the timeout elapses (event "timeout"). If another wait_for_event starts while one is pending, the older call returns event "superseded". Events that happen while nothing is waiting are queued and delivered in order on the next call, so nothing is missed. Once the session has ended every call returns "ended" immediately, until the user presses 録音再開 / Resume recording: recording then continues with the same transcript and timestamps keep counting up. The result always includes new_segments (transcript segments not yet returned to you, each {t0,t1,text} in seconds from recording start, plus "corrected" once you fixed that segment with correct_segments — "text" always stays the raw ASR output), total_segments, recording, ended, pending_question, pending_question_id and elapsed_s. Returns {error} until the user has started recording.

Page: /ramble-to-slides/

Input schema

{
  "type": "object",
  "properties": {
    "max_wait_s": {
      "type": "number",
      "description": "Maximum seconds to wait before returning event \"timeout\". Default 45, max 90."
    }
  }
}