“Known edge cases for the Whisper transcription model include misinterpreting silence as "thanks for watching," splitting single words into two chunks, and a lack of speaker recognition.”