Anyone done any work on using a model for transcription on the local device using the ANE? I've heard it kills the battery. Having to transcribe voice in the cloud is a serious impediment to end to end encryption for certain applications.
This is close:
https://github.com/ggerganov/whisper.cpp