Efficient Region-Aware Neural Radiance Fields for High-Fidelity Talking Portrait Synthesis | IEEE Conference Publication | IEEE Xplore