Ability to connect with custom speech-to-text and text-to-speech model or service using grpc or websocket and API Deepgram currently not support vietnamese, whisper has high latency. I have a local Vietnamese model which is more accurate and faster than whisper