So, what am I looking at? It seems like you fed the audio in PocketSphinx to get time-tagged text and the site basically shows said text as subtitles to what was said, is that the gist of it?
Yes, it is.
I'd like to improve the speech recognition and expected some advice about that.
Another possibility is to add a semantic level with NLP or use another library like Kaldi (http://kaldi-asr.org/).
Another particularity: the WAV file is serialized in JSON (as an array).