Whispering Context: Distilling Syntax and Semantics for Long Speech Transcripts