speech (1st sentence, 2nd paragraph).
2. Problem: Non-verbal communication in virtual environments is either manually controlled or non existent. Automatically generate gestures instead.
3. Key Idea: Separate gesture kinematics from shape and learn each separately to decrease sensitivity to noisy data.
4.a What the paper does: Generates avatar gesture animations from live speed using learned models. The paper also shows the flexibility of the models by using different sets of animations for learning and synthesis.
4.b. What it could be used for: This could be used for generating realistic conversational gestures in crowd simulation environments such as VR and games.
5. Resources: video