Not FunctionGemma related, but would love to see an open weights model from Google for speech to text transcription (diarization, timestamps, etc.).

Whisper is old and resource intensive for the accuracy it provides.