Definitely a hard problem. I was looking to leverage any existing API I can use. I've tried Google Vision for text and it's amazing. If there was something for audio...

I told a friend of mine who's into machine learning and AI.

I think that it's possible for the program to get to know how you pronounce things via machine learning. I'm not an expert in this space, though.

I could start with learning English, since from my experience software often works better than with other languages.