No textual conflicts. One rate-and-pitch persistence mechanism, not two:
main's profile sync (71a804f3, the access method, motor profile, vocabulary
level and voice, once the saved state is in) already saves the voice, so
this branch's own voice-only effect is dropped. The README's voice line now
describes the simpler rule (the best device voice speaks; robotic ones last).
Co-Authored-By: Claude Opus 5.5 <[email protected]>