Smooth Operators: LLMs Translating Imperfect Hints into Disfluency-Rich Transcripts