The technology is 100+ years old and has been used for 80 on human brain waves.
Almost 20 years ago, work at Radford was able to guess with 70 to 80 percent accuracy which of three possibilities within three parameters (size, shape and color) was being looked at, or being imagined with and without there being an attempt to verbalize it. They used a standard 16 channel external EEG. And a dozen different subjects.
Which "speech center(s)"? There's two main regions, neither of which can do the job alone. There's the areas where the material to be translated into speech get placed, and they can be read without having to try to work around linguistic encoding. Then there's people who lose their entire speech area, but come out being able to speak anyway because of backup/trainable areas taking over the job, or simply doing it in parallel all along.
You've got to have a damn good reason to carve open a skull. Surgical correction for epilepsy is a good reason, but the brain being tested before and after the surgery is hardly one to draw generalizations from. Given that previous work bested this without cutting into anyone, this is a dead end stunt.
There is also existing technology that would do the vocalizing job, also without surgery. Adapting it to an input based on a neural net 'best guess' output after training on an individual would be trivial compared to cutting open heads. Millions of people have heard it work, on a Pink Floyd album: "For millions of years mankind lived just like the animals. Then something happened to unleash the powers of his imagination -- he learned to talk." Many millions more have heard the same person/voice narrating the video version of his book "A Brief History Of Time".
TFA is some scary shit. With all the alternatives available, safer, better AND cheaper, there's no reason to do stuff like this, and none at all to suggest that it should be used as a basis to develop a technology.