Voiced-Aware Style Extraction and Style Direction Adjustment for Expressive Text-to-Speech