Text this: Multi-Level Feature Dynamic Fusion Neural Radiance Fields for Audio-Driven Talking Head Generation