I don't understand this part. Are they trying to pull the audio of the words out of the brain or something? I'd think it would be easier to use a dictionary of words, and use some machine learning to try and pull out the most likely next word from the brain activity, in which case 100% of the words would be intelligible
Evolving descriptive text of mental content from human brain activity
https://www.bbc.com/future/article/20260226-how-ai-can-read-your-thoughts