Whisper-level transcription accuracy should be sufficient