It's using OpenAI's latest audio model