The questions then become --- can you trust the system on-device transcription to preserve privacy or can you justify the size of the datasets for transcription if you've got to include it/download it for secure offline use; and can you make it work reasonably well in the languages your users use. (bonus points if it can reasonably detect the spoken language which may not match the written language of the phone).

I have been using on device transcription on my old Pixel. Here is a demo of the 2 years old technology: https://youtu.be/zfOPF069Xn0

But what I'm looking forward is the new technology in the Pixel 6. Besides much better transcription, caption and translation (see the video completely) it could even solve the real problem by letting people send text messages by speaking into theirs phones with better quality than before.

https://youtu.be/tSAYtwTqppA