Speech Recognition?! Technically feasible, but won't happen
Sorry, the Star Trek day when everything can be reliably controlled by voice alone will not arrive.
Even if somebody puts together the hardware and software that will reliably transcribe all accents for a given language there are other reasons why SR will not replace mechanical interactions. We still need feedback in some form to verify what was spoken has been properly understood - OK that could synthesised speech repeating back what the computer thinks you said. Also, once an individual sees/hears that their command has been incorrectly interpreted there is an overwhelming urge to use a mechanical interaction (keyboard, mouse gesture, touch screen) to correct or cancel it rather than risk trying to so using another voice command.
For me the killer technical issue is that for a computer to reliably differentiate between speech that is a command it should execute, and everything else we and others around us say, we would have to adopt a highly specific vocabulary for addressing the computer.
There is a bigger problem though that this would still not overcome. Sci Fi generally depicts speech control as an individual using their natural language to address a computer, sometimes prefixing what they say with a 'magic word'. We all understand what is going on because we are capable of analysing not just what the person is saying, but the entire context within which it is said. Thus we are able to determine not only what is being said but also the work out who or what it is being said to.
Imagine closing your eyes, listening to casual conversation between a group of people and identifying when you are being addressed, and when you are not. Sometimes you'll know because your name is said first. At other times you might be able to work it out from the context of the prior conversation. Sometimes though you'll not be able to know with a high degree of confidence if a question is aimed at you or not. It's the same for a computer - with only one source of information it won't be able to distinguish reliably enough when it is being addressed and when it is not. The irritation of false negatives (commands ignored) risk of false positives (non-commands treated as commands) will remain too high to adopt voice control in any situation where there are multiple people and/or multiple voice activated systems and/or anything vaguely safety critical. You can argue that this will improve over time but what rate is tolerable to joe public? And that's why it won't fly.