Speech, gestures, touch, controllers, keyboards, mice, trackpads, trackballs, everything else in the giant spectrum of input devices, all have different pros/cons and different accessibility. A failure often in sci-fi virtual reality is that everyone magically switches to a single mode of input, when even computing today suggests that every mode of input may be welcome (or even necessary) to users. VR/AR doesn't shift those accessibility needs as some imagine.
The interesting thing is that AR/MR suggest new input modes as well. There's a reverse skeumorphism that becomes more apparent in AR, along the lines of: in real life to accomplish a task I might turn some knob and that knob has seen decades of being the right tool for that task, why can't the AR respond to that knob as well. As AR/MR become more common and you start to see more "reality" in computing bleeding in from the other side, digital objects responding to physical ones, I think you'll start to see an expansion of what it even means to provide input to a machine. You might think of that as gesture controls, at least at first glance and some that may simply be advanced enough gesture recognition, but I think there's going to be a lot of bleed-over into the IoT space for custom controllers and sensors and touch surfaces. Some of those are going to look like existing physical objects, but extending their footprints deeper into "digital spaces" (many already are to one extent or another: your car's steering wheel, as one among many examples, may already have a digital footprint of some sort). I wouldn't be surprised if some of that leads to entirely novel seeming input devices in specific subdomains of human behavior.
