This repository provides a PyTorch implementation of a neural radiance field system for audio-driven talking head synthesis. The system trains separate NeRF models for head and torso regions, conditioning the generation on audio features to produce synchronized talking head videos. It was published at ICCV 2021.