Dunlin wavelet Nicolas Lecomte is acknowledged.
A new advanced machine learning AI tool has been developed by researchers to assist ecologists in training bird identification tools and monitoring rare species in their natural habitats. The tool is designed to generate lifelike bird songs, as reported in Techniques in Ecology and Evolution.
The availability of numerous phone applications and software options has simplified the process of identifying common bird species through their songs for ecologists and the general public. However, challenges arise when the recognition software has limited access to recordings or encounters unfamiliar bird species. This poses a particular challenge for conservationists monitoring rare bird species worldwide.
To address this issue, researchers at the University of Moncton in Canada have introduced ECOGEN, a groundbreaking deep learning tool that creates realistic bird sounds to enhance the representation of various species. These synthesized sounds can be utilized to train sound identification tools for ecological monitoring, which often lack sufficient data on rare species.
The study revealed that the artificial bird song samples generated by ECOGEN improved the accuracy of bird song identification by an average of 12%.
Ecologists utilize audiomoth sound monitoring devices to track wildlife, with Nicolas Lecomte being credited for the authenticity of this approach.
Dr. Nicolas Lecomte emphasized the urgent need for automated tools like sound monitoring to document changes in biodiversity resulting from significant global shifts in animal populations. However, there is a lack of comprehensive reference materials for the AI models used in identifying species through acoustic monitoring.
ECOGEN bridges this gap by creating new bird sound examples to support AI models, thereby expanding the existing library without disturbing the animals or requiring additional fieldwork, especially for species with limited recorded sounds, such as rare, elusive, or sensitive species.
The researchers suggest that synthesizing bird songs in this manner can aid in the conservation of threatened bird species while providing valuable insights into their vocalizations, behaviors, and habitat preferences.
Furthermore, the ECOGEN tool could have broader applications, potentially benefiting extremely endangered species like the regent honeyeaters, which face extinction due to the scarcity of adult birds to model their songs for the younger generation.
Dr. Lecomte highlighted the tool’s versatility, noting its potential application not only to birds but also to other animals like fish, insects, and frogs.
In addition to its adaptability, the open-source nature of the ECOGEN tool makes it easily accessible and usable on basic computers.
To enrich data for rare species with limited recordings, ECOGEN converts real bird track recordings into spectrograms (visual representations of sound) and generates new AI images based on these spectrograms. The resulting spectrogram images are then utilized to train animal sound recognition systems. The researchers utilized a database of 23,784 exotic bird recordings from 264 species worldwide for this study.
