What causes "Creating new recognition instance due to error" in moeru-ai/airi?

Usually self-healing — confirm the next log line says the new instance started successfully.

Creating new recognition instance due to error

Error message

Creating new recognition instance due to error

What it means

After an unclassified startRecognition() failure (not 'already started', not permission-related), the provider logs this warning and attempts recovery by discarding the current SpeechRecognition object and building a fresh one via createAndStartNewRecognitionInstance(). It is the mid-path of the restart strategy, not a terminal error — terminal failure is the subsequent 'failed to start after retry' path.

Source

Thrown at packages/stage-ui/src/libs/providers/providers/browser-web-speech-api/provider.ts:402

      if (errorMessage.includes('already') || errorMessage.includes('started')) {
        // Recognition is already running, this is OK
        console.info('Web Speech API recognition already running')
        return true
      }

      if (errorMessage.includes('not-allowed') || errorMessage.includes('permission')) {
        // Permission denied - user needs to grant microphone access
        const err = new Error('Microphone permission denied. Please grant microphone access and try again.')
        console.error('Web Speech API: Microphone permission denied')
        fullStreamCtrl?.error(err)
        textStreamCtrl?.error(err)
        deferredText.reject(err)
        deferredText.isRejected = true
        return false
      }

      // For other errors, try creating a new instance
      console.warn('Creating new recognition instance due to error')
      try {
        createAndStartNewRecognitionInstance(recognition)
        console.info('Web Speech API recognition restarted successfully with new instance')
        return true
      }
      catch (restartError: any) {
        const err = new Error(`Failed to start Web Speech API recognition: ${restartError?.message || String(restartError)}`)
        fullStreamCtrl?.error(err)
        textStreamCtrl?.error(err)
        deferredText.reject(err)
        deferredText.isRejected = true
        console.error('Web Speech API recognition failed to start after retry:', restartError)
        return false
      }
    }
  }

  // Add event listeners for debugging before starting

View on GitHub (pinned to 677329427f)

Solutions

Usually self-healing — confirm the next log line says the new instance started successfully.
If it recurs in a loop, capture the original errorMessage from the preceding 'start failed' log to identify the true cause.
Add a short cooldown or max-restart counter so a permanently failing engine does not recreate instances in a tight loop.
Check OS/browser speech service health and microphone device state when retries keep being needed.

Defensive patterns

Strategy: fallback

Try / catch

try {
  createAndStartNewRecognitionInstance(recognition)
}
catch (restartError: any) {
  const err = new Error(`Failed to start Web Speech API recognition: ${errorMessageFrom(restartError)}`)
  // fail the stream once; surface to user rather than looping retries
}

Prevention

Cap instance-recreation attempts to avoid restart loops.
Log the original start error alongside the recreation attempt for diagnosis.
Cool down between recreation attempts (the provider already delays restarts).

When it happens

Trigger: recognition.start() threw an error whose message matched neither the 'already started' nor 'not-allowed/permission' branches — e.g. InvalidStateError variants, 'service-not-allowed', or engine-specific exceptions — so the code recreates the instance to reset its lifecycle.

Common situations: Browser engine glitch leaving recognition in a bad internal state; switching input devices mid-session; speech service temporarily unavailable; long sessions where the old object stops accepting start().

Related errors

AI-assisted analysis of moeru-ai/airi@677329427f (2026-08-18). Data as JSON: /api/errors/b373c7a49b4ff807. Report an issue: GitHub.