In the last article I mentioned that I had scheduled a recording session for Tuesday morning, in order to properly capture my voice. The recording session could not have gone better. Ditto for the AI voice capture, which actually made me sound cooler and more sophisticated than in reality. Before I get to that I want to point out a few things I’ve learned from my prototype works in the meantime.
The original idea was to simulate a satirical TV talk show. Unfortunately, the AI is simply not good enough to reliably simulate crowd noise. It also requires a huge amount of audio editing to get audience sounds synced properly with the dialogue. I have neither the time nor the inclination to do that.
The new idea is basically the same, only instead of a TV talk show it’s a radio show, like fascist NPR. That works well with my voice, and allows me to naturally interject sounds as transition effects, such as the signal that we have cut to a recording that you hear in the below video. I think the result is fairly decent, although a lot of polishing work needs to be done. If you’re curious, yes I did have a great time creating and utilizing the terrible sound effects.
We’re still a ways away from the original implementation idea, which if you recall was based on the StarCraft video below. As I’ve said before, implementing what you see below is not overly difficult, but definitely comes with a lot of tedium, as I would have to manually mark the time each speaker starts, in addition to generating various speaking images for each character, flipping between them appropriately, fitting the text on screen, and an additional variety of miscellaneous details.
On the other hand, I always have the fallback option of simply slapping some static meme over the video and calling it a day, perhaps with a progress bar and an option to speed up or slow down the video. I’ve been making extremely good progress on this, and should have something up and running by late tomorrow.
I’m starting to think that the simple solution may be the best.
