Iterative Text-based Editing of Talking-heads Using Neural Retargeting

Supplemental Materials

[back to index]

Comparison between different phoneme search and stitching methods

We compare results from Fried et al. [2019] with our full fast search and stitching method (denoted "Modified Fried et al."), Fried et al. [2019] with just the phoneme context expansion, as well as vanilla Fried et al. [2019]. Our fast search and stitching algorithm can correctly close the mouth at \m, \b, \p phonemes and the context expansion produces less jitter, especially with full sentence syntheses.

That's all, folks people
Modified Fried et al. Fried et al. with context expansion Fried et al.
The quick brown spider fox jumped over the lazy dog
Modified Fried et al. Fried et al. with context expansion Fried et al.
The museum hires musicians every evening. (full sentence synthesis)
Modified Fried et al. Fried et al. with context expansion Fried et al.
Alimony harms a divorced man's wealth. (full sentence synthesis)
Modified Fried et al. Fried et al. with context expansion Fried et al.