Sarab S. Sethi

This paper introduces an autonomous drone platform for ecoacoustic surveys that uses intermittent locomotion to collect audio data without disturbing wildlife, achieving comparable biodiversity results to traditional methods.

This paper presents a machine learning approach to classify bird flight patterns using inexpensive equipment, achieving species-level classification with moderate accuracy.

This study uses national-scale passive acoustic monitoring to track avian migration patterns and develop species distribution models for three passerine species.

This study evaluates UAV-based acoustic monitoring for biodiversity assessment, finding that sampling intensity affects results but not strategy, and highlights its utility for broad trends over exhaustive inventories.

This study shows that combining LiDAR and automated bioacoustic analysis can effectively monitor avian biodiversity along a coffee production intensity gradient.

This paper demonstrates that pretrained neural networks and unsupervised learning effectively analyze coral reef soundscapes for ecological insights without manual annotation.

The paper develops interpretable machine learning methods to classify land use from soundscape data, improving accuracy by filtering non-informative audio and identifying key acoustic elements.

The paper applies a pre-trained bird vocalisation detection model to global audio datasets, achieving high precision for many species and enabling biodiversity monitoring across multiple spatial and temporal scales.

The paper introduces MAARU, a low-cost, open-source multichannel acoustic recorder for spatial ecosystem monitoring, demonstrating its accuracy in localizing sounds and improving species identification.

The study examines how individual species respond to tropical forest degradation, finding significant variation in responses across populations and surveys.

This paper develops ecoVAD, a VAD model for eco-acoustic data that outperforms existing models and enables privacy protection and human disturbance quantification.

The paper introduces catch22, a set of 22 time-series features that balance computational efficiency and classification performance across diverse biomedical and other time-series datasets.

The paper examines the relationship between structural connectome topology and regional BOLD signal dynamics in the mouse brain, finding strong correlations between weighted in-degree and fMRI time series autocorrelation.