Function startRecordingSession

src/hooks/useVoice.ts:633–1011

()

  // ── Start a new recording session (voice_stream connect + audio) ──
  async function startRecordingSession(): Promise<void> {
    if (!voiceModule) {
      onErrorRef.current?.(
        'Voice module not loaded yet. Try again in a moment.',
      )
      return
    }

    // Transition to 'recording' synchronously, BEFORE any await. Callers
    // read state synchronously right after `void startRecordingSession()`:
    // - useVoiceIntegration.tsx space-hold guard reads voiceState from the
    //   store immediately — if it sees 'idle' it clears isSpaceHoldActiveRef
    //   and space auto-repeat leaks into the text input (100% repro)
    // - handleKeyEvent's `currentState === 'idle'` re-entry check below
    // If an await runs first, both see stale 'idle'. See PR #20873 review.
    updateState('recording')
    recordingStartRef.current = Date.now()
    accumulatedRef.current = ''
    seenRepeatRef.current = false
    hasAudioSignalRef.current = false
    retryUsedRef.current = false
    silentDropRetriedRef.current = false
    fullAudioRef.current = []
    focusFlushedCharsRef.current = 0
    everConnectedRef.current = false
    const myGen = ++sessionGenRef.current

    // ── Pre-check: can we actually record audio? ──────────────
    const availability = await voiceModule.checkRecordingAvailability()
    if (!availability.available) {
      logForDebugging(
        `[voice] Recording not available: ${availability.reason ?? 'unknown'}`,
      )
      onErrorRef.current?.(
        availability.reason ?? 'Audio recording is not available.',
      )
      cleanup()
      updateState('idle')
      return
    }

    logForDebugging(
      '[voice] Starting recording session, connecting voice stream',
    )
    // Clear any previous error
    setVoiceState(prev => {
      if (!prev.voiceError) return prev
      return { ...prev, voiceError: null }
    })

    // Buffer audio chunks while the WebSocket connects. Once the connection
    // is ready (onReady fires), buffered chunks are flushed and subsequent
    // chunks are sent directly.
    const audioBuffer: Buffer[] = []

    // Start recording IMMEDIATELY — audio is buffered until the WebSocket
    // opens, eliminating the 1-2s latency from waiting for OAuth + WS connect.
    logForDebugging(