Department of Defense Fiscal Year (FY) 2010 Budget Estimates May 2009, USA
``` Silent Talk (U) Silent Talk will allow user-to-user communication on the battlefield without the use of vocalized speech through analysis of neural signals. The brain generates word-specific signals prior to sending electrical impulses to the vocal cords. These signals of “intended speech” will be analyzed and translated into distinct words, allowing covert person-to-person communication. This program has three major goals: a) to attempt to identify electroencephalography patterns unique to individual words, b) ensure that those patterns are generalizable across users in order to prevent extensive device training, and c) construct a fieldable pre-prototype that would decode the signal and transmit over a limited range. ``` https://commons.wikimedia.org/wiki/File:Fiscal_Year_2010_DAR...
This looks no harder than to train custom Kaldi (circa 2017) phoneme model on brain waves, and using remaining Kaldi's functionality for everything else, except for text-to-speech. There was WaveNet for the TTS at that time, with sound quality that is good enough for (and can be improved by) radio transmission.
Thanks for a link!