After several weeks of development, Cheshire has made great strides in its progress. A major step came when I stopped trying to roll my own Python mapping for PocketSphinx, and chose to simply start using the existing mapping available through pip. Turns out all the functionality I required was already available in this package, and running through the documentation provided all the guidance needed.
At this point, we have a functioning prototype with a web interface that takes a provided audio file, breaks it down, reviews the predicted text in each chunk, and accepts a text transcription before retrieving the audio timings. Once all of the audio has been reviewed, a preview of the animated audio timings is provided, and the final data can be exported.
Now that an actual prototype for extracting timing data is working, we have begun work on a plug-in so that this timing data can be applied for generating actual animations. At the moment, our platform of choice for this is the Godot game engine. While Mad Wonder is relatively new to Godot tool development, we’ve already enjoyed some success at the process, and are finding it relatively easy to get into. I am hopeful that we will have an animated preview to show off some time in the coming month.