Sound Array
Concept
Spatial audio analysis on a microphone array: capture multi-channel audio, estimate time-of-arrival differences between channels to localise sound sources, apply beamforming to isolate them directionally, and classify each isolated stream — vehicles, aircraft, wildlife — into provenance-bearing events that an analysis agent can act on.
The pipeline is built around a single consistency boundary: a windowed audio snapshot becomes the input to every downstream estimator, so localisation, beamforming, and classification all reason about the same coherent slice of time.
Architecture
multi-channel USB/HAT"] SNAP["Snapshot Window
consistency boundary · happens-before"] GCC["GCC-PHAT
per-pair delay estimation"] BF["Beamforming
geometry-driven directional isolation"] CLS["Gated Classification
vehicles · aircraft · wildlife"] EV["Events
provenance-bearing predicates over time"] MIC --> SNAP SNAP --> GCC GCC --> BF BF --> CLS CLS --> EV
Each stage is a deterministic transform over the snapshot: the same input window produces byte-identical localisation, beamforming, and classification outputs. Geometry provenance and timing certainty propagate downstream, so an event carries the uncertainty of its inputs rather than absorbing it silently.