4 ms·
Looks pretty good. Their example video sounds terrible and robotic, seemingly by design. Would the best approach to actually deceiving the viewer be a voice im
by np_tedious 9y ago
Looks pretty good. Their example video sounds terrible and robotic, seemingly by design.
Would the best approach to actually deceiving the viewer be a voice impressionist paired with this technique?
- save_ferris 9y agoWith the rate at which this technology is advancing, I see this as more of an MVP to a much more powerful tool in the not-too-distant future. But I also wouldn't hate being wrong on this one.
- gwern 9y agoThe best approach would be (only half-joking here) to wait a year and then use Baidu or a dozen other (Lyrebird included) group's voice synthesis neural nets (generally either a RNN or a WaveNet-style CNN). The last WaveNet samples were indistinguishable from real for me.