Sound Event Localization And Detection

A Sound Event Localization and Detection (SELD) system takes multi-channel audio input and outputs the time-activity trajectories of target sound classes along with multiple corresponding spatial trajectories. This system characterizes acoustic scenes through spatiotemporal features and is applicable to various machine cognition tasks such as environment type inference, self-localization, navigation without vision, specific sound source tracking, smart home applications, scene visualization systems, and audio surveillance.