Towards Conversational Medical AI with Eyes, Ears and a Voice