Abstract
1. Passive acoustic monitoring (PAM) coupled with artificial intelligence(AI) is becoming an essential tool for biodiversity monitoring. Traditional PAMsystems require manual data offloading and impose substantial demands onstorage and computing infrastructure. The combination of on-device AI-basedprocessing and network connectivity enables local data analysis andtransmission of only relevant information, greatly reducing storage needs.However, programming these devices for robust operation is challenging,requiring expertise in embedded systems and software engineering. Despite theincrease in AI-based models for bioacoustics, their full potential remainsunrealized without accessible tools to deploy them on custom hardware andtailor device behaviour to specific monitoring goals. 2. To address thischallenge, we develop acoupi, an open-source Python framework that simplifiesthe creation and deployment of smart bioacoustic devices. acoupi integratesaudio recording, AI-based data processing, data management, and real-timewireless messaging into a unified and configurable framework. By modularisingkey elements of the bioacoustic monitoring workflow, acoupi allows users toeasily customise, extend, or select specific components to fit their uniquemonitoring needs. 3. We demonstrate the flexibility of acoupi by integratingtwo bioacoustic classifiers: BirdNET, for the classification of bird species,and BatDetect2, for the classification of UK bat species. We test thereliability of acoupi over a month-long deployment of two acoupi-powereddevices in a UK urban park. 4. acoupi can be deployed on low-cost hardware suchas the Raspberry Pi and can be customised for various applications. acoupistandardised framework and simplified tools facilitate the adoption ofAI-powered PAM systems for researchers and conservationists. acoupi is onGitHub at https://github.com/acoupi/acoupi.