The Sound of Healthcare: Improving Medical Transcription ASR Accuracy with Large Language Models