Gesture segmentation was performed as described in:
Ferstl, Y., Neff, M., & McDonnell, R. (2020). Adversarial gesture generation with realistic gesture phasing. Computers & Graphics, 89, 117-130.

That is, 4 hours of our 2nd dataset (Trinity Speech-Gesture II) were hand-annotated; then, we trained an classifier network and automatically annotated the remaining data (including our 1st dataset (Trinity Speech-Gesture I)).
The resulting segementations can be expected to have some errors; for a discussion please refer to the work noted above.

If you use this gesture segementation, please cite this work. The following BibTeX should be used:

@article{ferstl2020adversarial,
  title={Adversarial gesture generation with realistic gesture phasing},
  author={Ferstl, Ylva and Neff, Michael and McDonnell, Rachel},
  journal={Computers \& Graphics},
  volume={89},
  pages={117--130},
  year={2020},
  publisher={Elsevier}
}