Moreover, speech recognition isn't a product you can't launch in a faulty state. You can sell a bad product, make a profit, and iterate without any consequence.
It's not the same with self driving cars, which won't be profitable till you get rid of the driver entirely. And you can't launch a ok product and iterate. It has to be almost perfect when it comes to safety.
Barring geofencing or legislative changes, you have to get it in a perfect state before you roll it out.
When the VC money dies down, and when such big companies are no longer optimistic at having a viable self driving car, the research is going to slow down drastically, till a new wave of funding comes along from some other boom to resuscitate the research.