How is the temporal alignment achieved?

[sukun1045]

Hi, thanks for sharing the code! I have read the papers and watched the demo video. I find that the generated sound does temporally align with the visual information but I am not sure how you achieve that since I haven't found any component in your paper explicitly designed for that purpose. Therefore, I am curious about how the RegNet is able to guarantee temporal alignment. Thank you.