The sound of hope: searching for critically endangered species using acoustic template matching

Bioacoustics, Journal Year: 2023, Volume and Issue: 32(6), P. 708 - 723

Published: Oct. 23, 2023

Passive acoustic monitoring (PAM) has become increasingly popular in biodiversity. It produces large amounts of data and can provide a foundation for understanding the long-term consequences environmental degradation. However, extracting biological information from such extensive datasets be challenging requires advanced computational skills. Herein, we introduce streamlined workflow detecting signals three critically endangered birds: Cherry-throated Tanager (Nemosia rourei), Alagoas Antwren (Myrmotherula snowi), Blue-eyed Ground-dove (Columbina cyanopis). As these species are among world's most birds, locating new populations is top priority. We chose potential templates based on parameters vocal repertoire evaluated their performance using soundscapes with known composition (gold standard data). To evaluate efficiency templates, used precision recall metrics found that achieving high rates comes at cost rates. Although gold to calibrate our algorithm, large-scale validations have revealed limitations as some exhibited significantly lower values. The use binomial models helped reset values 90%. Our process efficiently, helping monitor species, locate population dynamics.

Language: Английский

---

A Review of Automated Bioacoustics and General Acoustics Classification Research

Sensors, Journal Year: 2022, Volume and Issue: 22(21), P. 8361 - 8361

Published: Oct. 31, 2022

Automated bioacoustics classification has received increasing attention from the research community in recent years due its cross-disciplinary nature and diverse application. Applications range smart acoustic sensor networks that investigate effects of vocalizations on species to context-aware edge devices anticipate changes their environment adapt sensing processing accordingly. The described here is an in-depth survey current state monitoring. examines alongside general acoustics provide a representative picture landscape. reviewed 124 studies spanning eight research. identifies key application areas techniques used audio transformation feature extraction. also algorithms systems. Lastly, challenges, possible opportunities, future directions bioacoustics.

Language: Английский

---

Grad-CAM++ is Equivalent to Grad-CAM With Positive Gradients

Published: Aug. 29, 2022

The Grad-CAM algorithm provides a way to identify what parts of an image contribute most the output classifier deep network. is simple and widely used for localization objects in image, although some researchers have point out its limitations, proposed various alternatives. One them Grad-CAM++, that according authors can provide better visual explanations network predictions, does job at locating even occurrences multiple object instances single image. Here we show Grad-CAM++ practically equivalent very variation which gradients are replaced with positive gradients.

Language: Английский

---

Who is calling? Optimizing source identification from marmoset vocalizations with hierarchical machine learning classifiers

Journal of The Royal Society Interface, Journal Year: 2023, Volume and Issue: 20(207)

Published: Oct. 1, 2023

With their highly social nature and complex vocal communication system, marmosets are important models for comparative studies of and, eventually, language evolution. However, our knowledge about marmoset vocalizations predominantly originates from playback or interactions between dyads, there is a need to move towards studying group-level dynamics. Efficient source identification essential this challenge, machine learning algorithms (MLAs) can aid it. Here we built pipeline capable plentiful feature extraction, meaningful selection, supervised classification up 18 marmosets. We optimized the classifier by building hierarchical MLA that first learned determine sex source, narrowed down possible individuals based on then determined identity. were able correctly identify individual with high precisions (87.21%-94.42%, depending call type, 97.79% after removal twins dataset). also examine robustness across varying sample sizes. Our promising tool not only but analysing other species.

Language: Английский

---

Improving Misfire Fault Diagnosis with Cascading Architectures via Acoustic Vehicle Characterization

Sensors, Journal Year: 2022, Volume and Issue: 22(20), P. 7736 - 7736

Published: Oct. 12, 2022

In a world dependent on road-based transportation, it is essential to understand automobiles. We propose an acoustic road vehicle characterization system as integrated approach for using sound captured by mobile devices enhance transparency and understanding of vehicles their condition non-expert users. develop implement novel deep learning cascading architectures, which we define conditional, multi-level networks that process raw audio extract highly granular insights understanding. To showcase the viability build multi-task convolutional neural network predicts cascades attributes misfire fault detection. train test these models synthesized dataset reflecting more than 40 hours augmented audio. Through fuel type, engine configuration, cylinder count aspiration type attributes, our CNN achieves 87.0% set accuracy detection demonstrates margins 8.0% 1.7% over naïve parallel baselines. explore experimental studies focused features, data augmentation, reliability. Finally, conclude with discussion broader implications, future directions, application areas this work.

Language: Английский

---