SoundscapeAgent: Agentic Soundscape Construction for Controllable Synthesis and Scalable Audio-Language Supervision

Researchers introduced SoundscapeAgent, an LLM-driven framework for controllable soundscape synthesis.

It turns user intent into executable scene plans for multi-event audio generation, asset retrieval or creation, and temporal arrangement.

The system also supports human-in-the-loop editing and aligned metadata export for more inspectable audio creation.

Experiments show strong synthesis quality and that agent-generated data improves downstream audio reasoning.