A million orthopteran singers from a year of continuously operating field recorders detected, tracked and organized by song
En palabras de los autores
Continuous acoustic monitoring has become an invaluable tool in ecology and conservation biology, with hundreds of thousands of automated recording units deployed worldwide. While semi-manual annotation is adequate for some uses, fully automated methods to segregate and identify the source of animal sounds are essential to utilize these data to their full extent. Significant progress has been made for birds, whose calls are generally discrete and structured, and where large annotated reference libraries are available. However these methods do not extend neatly to insects, whose choruses dominate nighttime, and often daytime, soundscapes across most of the world. While many approaches have been proposed to fill this gap, an efficient end-to-end system for annotating insect calls that can operate at the scale of available data has not emerged. Here we describe an unsupervised method that uses the intrinsic, highly repetitive structure of the songs of many insects, especially orthopterans, to identify and separate distinct songs in narrow (minute) time windows, connect them across windows for as long as entire nights as their carrier frequencies and syllable repeat rates vary with time, temperature and other factors, and assemble them across days and locations into a catalog of singers and song types across time, space and conditions. We apply this method to thirteen months (November 2024 to November 2025) of recordings from 27 autonomous recording units in different microhabitats at a field site in southwestern Arkansas, 175,896 recorder-hours in all. We detected 10.9 million singers in five-minute windows and linked them, within each night or day, into 911,005 tracks, of which 103,802 lasted two hours or more (71,770 at night, 32,032 by day), a total of 335,882 hours of song. In the plane of carrier frequency against syllable repeat rate these singers fall on discrete lines along which each slides with temperature, and we use this organization to define 45 song types, 37 trilling and 8 chirping, each composed of dozens to thousands of singer-nights. Within singers, syllable rate rises by about 3 percent per degree and carrier frequency by about 1.5 percent in nearly every type, recapitulating and extending classical observations on the temperature dependence of orthopteran song. Published and reference recordings placed on the same plane fall on these lines, and species of Anaxipha recorded by Walker and Funk land on single types at the right temperatures. More generally, this method and the data it produces provide a robust means of studying changes in insect song over time and space, and of understanding the behavioral and physical forces that shape its production.
En palabras simples Resumen con IA
Los investigadores desarrollaron un método computacional que encuentra cantos de insectos en grabaciones sin necesidad de etiquetas humanas. Estos cantantes se agruparon en 911,005 pistas, de las cuales 103,802 duraron dos horas o más. Los cantos cayeron en 45 tipos según su tono y qué tan rápido se repiten las sílabas, con cada tipo conteniendo muchas noches-cantante. El método puede ayudar a estudiar cómo cambian los cantos de insectos con el tiempo y el espacio.
Escrito por Kimi K2.5 usando solo el abstract de arriba. Jev verificó que cada oración esté respaldada por las oraciones del abstract que cita y eliminó las que no. Pasá el cursor por una oración para ver de dónde sale. Se eliminó 1 oración por no estar respaldada por el abstract.
Apareció: jueves, 22 de octubre. bioRxiv. Preprint, todavía sin revisión por pares.