Speech recognition had got quite good on mobile. How?
Not perfect, but good to the extent that I use it a fair amount on mobile to speed up my text input by using speech-to-text in WhatsApp, SMS, Gmail, etc. I voice-type and then edit to correct the errors.
My guess, lots of user data + user corrections. When we correct what our phones mishear, that data helps train the speech recognition models to get more accurate over time.
I’m particularly impressed by speech recognition on Apple Watch. On device, via tiny microphones, in noisy environments, and still works despite my non-native accent.