Oi David. Challenge for you. Can you map a pre-made vocal program to a text output that can be attached to a neural network, preferably bilingual, and find a way to have that NN run constantly unless turned off in a way that simulates a chatbox, with the audio attached obviously, unless audio is muted via button? 
Further, can you figure out a way to attach a user and password to it so that it responds differently based on whether someone has those?
I can provide the voice banks, and this is unrelated to our earlier conversation, I'm thinking of making something for an offline friend and this would signifigantly speed up the process. Even just a step by step would be useful.