INSTINCT: The infrastructure for noise and soundscape tolerant investigation of nonspecific call types

Abstract

INSTINCT is open-source, command line software for custom data pipelines, developed by the NOAA’s Alaska Fisheries Science Center to formalize AI workflows for passive acoustic monitoring (PAM) of sounds produced by marine mammals and to provide a framework for new algorithm development. INSTINCT is built in Python over the Luigi framework and was designed to be lightweight and Window/Linux compatible. To promote collaboration, the software is partitioned into a core module and organization-governed submodules. The original INSTINCT detection and classification algorithm has been applied to various Alaska region cetacean species and presents a powerful CPU-based approach for those seeking to train, evaluate, and deploy a generalized detector. Furthermore, INSTINCT pipelines are in development to train and deploy GPU-based algorithms, structure annotation workflows, and pioneer complex workflows. Case studies will be presented to highlight the existing capabilities of the software and new workflows that are currently being explored. This talk will also cover design philosophy, technical specifications, and instructions for implementation. Although INSTINCT continues to progress at AFSC, it will benefit greatly from a larger network of collaborators, and ideally, will give back to the community via innovation of new AI approaches in PAM.