This suggests future work on assistive devices that may be able to "speak for" a patient by ML and manual algorithmic adjustment... Huge quality-of-life improvement potential for people with limited/impaired vocal-ability.
Beyond that, there is far wider potential for translating brain scans directly and skipping vocal-aural pathway by communicating with others digitally.
I was thinking this as well. It reminded me of the "Commander Pike" character in the original Star Trek that could only respond yes or no through a light.
There is evidence that it is. They have been able to record images people are seeing from fMRI data. The information is there, it's just a matter of processing it.
Right now, researchers are putting together functional brain structure maps combining fMRI, SPECT and PET scans for Alzheimer's research. Apart from the inconvenience of wearing a giant magnet, pulling together one or more brain scan techs with compute resources might be a way to pull it off. Heck, if the compute power needed were too bulky to be practical, it might be possible to offload it to a hosted service. (Talking a solution 15-18 yrs out anyhow.). It's entirely reasonable that keyboards would be slower than thought input in 50 years.
Beyond that, there is far wider potential for translating brain scans directly and skipping vocal-aural pathway by communicating with others digitally.