A specialist that listens and reads simultaneously — this model fuses audio scene understanding with text into a shared embedding space, letting you compare sounds and descriptions directly. It excels at matching audio concepts to language without requiring actual audio at inference time, working purely from text descriptions of sounds. The trade-off is narrow focus: it's built for audio-language alignment tasks, not general-purpose text understanding.