Introducing frontier open source speech understanding models.