Next-generation audio models in the API