purely on-device voice to text by running local transformer models via FUTO Voice Input https://voiceinput.futo.org/