High-Fidelity and Freely Controllable Talking Head Video Generation