Realistic Speech-Driven Talking Video Generation with Personalized Pose

Table 3

Evaluation metrics used NME (%) on facial landmarks (lower is better).

	Orig.	Only-GRU	TTS-mel	Text
0.5	4.925	5.673	5.871	5.693
1.0	4.921	5.640	5.885	5.690
1.5	4.853	5.644	5.877	5.614
2.0	4.907	5.647	5.829	5.607