Today's Large Audio Language Models (LALMs) are stuck in an offline paradigm: you hand them a complete audio clip, wait, and get a reply. Streaming audio models exist, but each one only handles a ...