Comparative Analysis of Audio Feature Extraction for Real-Time Talking Portrait Synthesis