Voice typing is a cool feature, have you considered whisper wasm instead of OpenAI api?