Description
Source Motion with Sound
A standing-up motion with recording of my voice describing motion.
Target Motion
Another standing-up motion.
Synthesis Constraints
Automatically defined by motion matching algorithm.
Comments
The input soundtrack plays a different sound for sitting, standing-up and walking motions. Given a new standing-up motion, we want the same sounds sounds to play for the same motions. Our algorithm automatically defines constraints for the synthesis to achieve this effect.
The algorithm works by first segmenting the source and target motion into fragments. We obtain the 3D barycentre of the figure at each frame. We then detect the minima of the norm of the motion of the 3D barycentre. These define the segmentation boundaries of the fragments. This is done over both the source and target motions.
We then calculate the distance between each source fragment to every other target fragment. The top-3 best source matches for each target fragment are kept. We then find a path that maximizes the use of consecutive source fragments. The audio portion assigned to each source fragment is then assigned as a constraint to each target fragment. The synthesis finally synthesizes the resulting soundtrack.
Self-Similarity Synthesis
|
Source Motion with Sound
Target Motion
New Sound