Gemini 3.5 Transcribe as a dictation engine #653

Dictation already has Whisper / Parakeet / native, plus cloud options (OpenAI transcribe, ElevenLabs Scribe, Voxtral). Gemini is in the chat provider list, but it isn't an option for speech-to-text.

I'd like to use the Gemini API for dictation, specifically gemini-3.5-transcribe and gemini-3.5-transcribe-live. I've been using that model and the transcripts are cleaner than what I get from the local engines. Filler removal and cleanup come with the model, so it doesn't really need a separate rewrite pass.

The Gemini API key is already in settings for chat. Reusing it and adding those models to the dictation dropdown would be enough. Hold-to-speak can stay as it is: record while held, transcribe on release, or stream if the live endpoint is a better fit.

Happy to test a build.