There’s a problem with today’s voice recognition systems: They’re just too slow. Anyone who has waited in frustration while Siri or Google’s Voice Search “thinks” about even the simplest commands knows what I’m talking about. The problem isn’t voice recognition software per se, which is more accurate than ever. The problem is that voice recognition is still a challenging enough problem, computationally, that all the major consumer platforms that do it—whether built by Google, Apple or Microsoft with the new Xbox—must send a compressed recording of your voice to servers hundreds or thousands of miles away. There, computers more powerful than your phone or game console transform it into text or a command. It’s that round trip, especially on slower cellular connections, that make voice recognition on most devices so slow.