Automatic Quality Assessment of Speech-Driven Synthesized Gestures

Table 4

Evaluation of 3 models with data augmentation.

Model	Data augmentation	MSEtest	PCC	value	Error max	Error mean	Error std
Baseline	Padding with last frame	0.622	0.580		18.211	6.676	5.001
Structure improved	Padding with last frame	0.574	0.661		17.846	5.965	4.228
Algorithm adjustment	Padding with last frame	0.591	0.657		20.330	5.745	4.911
Baseline	Padding with zero	0.686	0.499		20.218	7.019	5.976
Structure improved	Padding with zero	0.617	0.567		19.130	6.333	5.494
Algorithm adjustment	Padding with zero	0.673	0.544		23.653	6.131	5.765