Towards Natural Gesture Synthesis: Evaluating gesture units in a data-driven approach to gesture synthesis

Videos complementing the IVA-07 submission (Paper 74)

For better playback it is best to first download the videos before playing them: right click the link and select "Save Target As..."

1. Animations recreated from the manual annotation

The following two clips demonstrate that our annotation scheme captures the shape and timing of the original gestures.

• video sample for speaker JL:    JL-recreated.mpg (13 MB)
• video sample for speaker MR:   MR-recreated.mpg (7 MB)

2. Animations automatically generated from text input

These videos show gestures that were automatically generated and animated from input text (marked-up as described in the paper). The first two clips show gestures generated from text transcripts of the samples shown above. The third clip shows gestures generated on novel text: a synthesized version of the Star Wars IV prologue.

• video sample with JL profile: JL-generated.mpg (11 MB)
• video sample with MR profile: MR-generated.mpg (12 MB)
• video sample (Star Wars) with JL/MR profiles side by side: SW-both.mpg (8 MB)

3. Videos used in the g-unit experiment

We replaced these files on May 16 with mpg-encoded files, just in case you could not read them before, please try again.

Four clips (out of 22) used in the g-unit experiment. The gestures in the "unit version" were automatically generated by our system, whereas in the "singleton version" the g-units were manually manipulated to make the gestures singletons.

• sample pair 1: unit version | singleton version (4 MB each)
• sample pair 2: unit version | singleton version (6 MB each)