Smoothen expressive avatar's transitions between words

[kevinjosethomas]

Description

In the current interface, the expressive avatar has no transition between words. As a result, it tends to teleport around word-to-word, making it significantly harder to understand. We can probably add a few keyframes between each word where the avatar moves its hands and body from one location to another.

I'm not sure how this would visually work. Maybe calculating the starting point and endpoint between each frame and simply inferencing frames in between could work? I'm not sure if it would lead to inhuman