Transcribe audio to text with speaker diarization