The article says, "But detecting sounds in a driver’s car without permission was clearly invasive." But it also explicitly says it would use the accelerometer and gyroscope. Perhaps these are sensitive enough to pick up vibrations from the speaker.

Different apps have unique vibration patterns, as well. A text message on my iPhone is two buzzes, but a NYTimes news alert is one. Perhaps if you detected a single buzz, and then picked up motion because I'm double-clicking the home button, you could presume that I'm switching apps to respond to the notification.

Most drivers would have sound on and not purely use vibrations, because the phone is not in their pocket.