Audient adds growing concept memory to LLM audio perception

The open-source framework Audient provides a local audio perception layer for LLM agents that continuously processes sound and builds a memory of concepts through use. It uses CLAP embeddings, Whisper, Silero VAD, and sqlite-vec to gate, fingerprint, and recognize audio events without requiring constant LLM calls.

• The system gates continuous audio streams using energy, novelty, and Silero VAD to identify candidate regions.
• It generates fingerprints via CLAP embeddings and symbolic features like spectral shape and temporal drift.
• A sqlite-vec index performs top-k cosine similarity against learned concepts, firing server-side matches when calibrated.
• Unrecognized events are queued for labeling, allowing new concepts to be added without retraining the model.

This approach allows agents to react to unexpected sounds like crashes or alarms by reasoning over symbolic layers rather than raw audio.