In the MeetingCopilot web app, you encountered an audio routing bug where two different AudioLines instances caused mismatched speaker lines. How did you identify and fix this issue?

During the real‑time transcription, I discovered that the transcript was out of sync because two separate AudioLines instances were processing the same audio stream. One instance was attached to the Whisper pipeline, while the other was bound to the UI component. The mismatch caused speaker lines to be duplicated or dropped. To fix this, I first instrumented the AudioLines constructor to log the instance ID and the stream it receives. I then refactored the code so that a single AudioLines instance is created at the top level and passed down via context to both the Whisper worker and the UI component. I also added a shared event emitter that broadcasts speaker change events, ensuring that the UI always receives the latest speaker line. After the refactor, the transcript stayed in sync, and the active speaker highlight behaved correctly. The key debugging steps were: isolate the source of duplication, unify the instance, and use a single source of truth for speaker state.