We compare to Neural Voice Puppetry [Thies et al. 2019] (ongoing concurrent work).
The user study shows that the difference between our results and NVP is not statistically significant.
Nevertheless, closely examining the videos generated by the two approaches, we find that our method often does a better job of closing
the mouth on \m, \b, and \p phonemes. We also note that while our user studies evaluate our automatic results, unlike NVP, our tool
also provides refinement and performance controls that can be used to improve results over the course of an interactive editing session.
Neural Voice Puppetry | Our result |
---|---|
Neural Voice Puppetry | Our result |
Neural Voice Puppetry | Our result |
Neural Voice Puppetry | Our result |
Neural Voice Puppetry | Our result |