While there seem to be plenty of reasons to doubt the conclusion of the post, let's assume it's right. To me, that would seem to mean that the current technology for voice recognition is based on a flawed assumption, and that progress will not be made until a new technology comes along that tosses out the old assumptions and starts fresh with a new look at the problem. Once a technology, even a flawed one, is established, it's devilishly hard to get rid of it and start from scratch. (We mostly all use QWERTY keyboards, which were actually designed to slow down typing, almost 50 years after the problem they were created to solve has vanished.) But sometimes that's what it takes to solve the problem.