Voice makes sense as a third input mode, but it seems like a poor replacement for the controls you need while stressed or time-constrained. In a car the real UX test isn't "can the system eventually understand me?", it's "can I do this eyes-free, in noise, with high confidence, in under a second?" That's why defrost, wipers, volume, and temperature feel fundamentally different from "navigate to X". Touch is fine for discoverability, voice is fine for occasional commands, but safety-critical/high-frequency actions still want dedicated hardware.