Settings

Theme

Using generative neural nets in keras to create ‘on-the-fly’ dialogue

neuralniche.com

3 points by hassiktir 10 years ago · 4 comments

Reader

strangecasts 10 years ago

How hard would it be to approximate the time/length of each word in the data set? I think you could make some quite interesting videos by matching the generated monologues to the original audio clips :)

  • hassiktirOP 10 years ago

    I'm not entirely sure what you mean, as in feed back the predictions for some length to see what the dialogue's become?

    • strangecasts 10 years ago

      Don't worry, it was just an amusing and honestly really impractical idea :)

      The idea would be to make an educated guess at where each word occurs in the video - going off the time and subtitle data from pysrt - and build a dict linking words to when they occur in the video. You could then use MoviePy and stitch together a video version of the generated dialogue, by looking up the appropriate clip for each word.

      • hassiktirOP 10 years ago

        ahh that does make sense now, and think it could be very feasible with a much more complex and sort of blended NN since the .SRT file's do have the time for each subtitle phrase (i.e.

          7
          00:00:23,060 --> 00:00:24,619
          give a turnaround version
        ) but i am not sure the best way to go about doing something like this.

Keyboard Shortcuts

j
Next item
k
Previous item
o / Enter
Open selected item
?
Show this help
Esc
Close modal / clear selection