Using the concept of visemes, I have mapped all the sound based pose curves for Metahuman face. Additionally different expressions are layered over it to have full modular control.
Listening and response is taken care by the backend technology and passes back JSON data to UnrealEngine based on which the entire speech set is dynamically played by the Metahuman mesh.
This enables, backend to have full control over time, speed, modulation,and voice for audio and animation