Pyweek 31 - postmorten on the speech to text experience
DR0ID on 2021/05/08 13:07
of GD
We expected to have some DNW due to our dependencies. We never had that many dependencies in a pyweek before. Our
experience told us, this will be trouble. And yes, it was trouble apparently. The strange thing is, that we didn't
hear about it. No one except two persons asked for help.
The frustrating thing about this type of interface is, that even if you have passed all technological difficulties,
it is no guarantee that it will work. It may very be that your accent is not recognized. Or your voice. Or
whatever causes it not to behave as expected. For me it worked pretty well, even I'm not a native english speaker.
But well, I trained a week long and I'm certainly biased. And native english speakers had trouble getting it
to recognize the commands. There are other models available but they are bigger (~1GB or more), but I don't know
if they would work better. Our game shipped with an already compressed 40MB file which is heavy enough!
To give you all the chance to play it through I made a version with a keyboard interface. It will behave the same
as if you would have spoken the words (just type the answer and submit it by pressing 'enter').
I hope you haven't read the spoilers yet since it would spoil the game. It also would be much appreciated if you could
give feedback as during the judgment period (Fun / Prod / Inno, 5 best - 1 worst).
Overall I found it an interesting pyweek. The theme was not our first choice and frankly, we didn't have a good idea.
And since quite some time I wanted to try text to speech. After some experiments I also found some libraries that
do the opposite: speech to text. Due to the asynchronous nature of those libraries the next step was to look into the
multiprocessing module of python. The result is this game. Its hard to close. On purpose. Unless you 'exterminate' it!