Temporal patterns in Malaysian rainforest soundscapes demonstrated using acoustic indices and deep embeddings trained on time-of-day estimationa).

This article focuses on the use of deep ecoacoustic embeddings, derived from a deep neural network trained on time-of-day estimation, to analyze temporal variations in tropical soundscapes and compares this approach with conventional ecoacoustic indices. Conducted on Penang Island’s tropical rainforests and urban green spaces, the study collected extensive passive acoustic monitoring data over 17 months using autonomous recording units. Results indicate that both conventional acoustic indices and deep ecoacoustic embeddings show comparable performance in capturing diel and seasonal soundscape patterns, with deep embeddings demonstrating superior accuracy in time-of-day prediction but slightly lower accuracy in habitat-type classification. The research highlights the potential of deep learning methods to enhance long-term, large-scale ecological monitoring by providing ecologically relevant acoustic features without requiring manual labeling, and suggests future directions for refining these models and validating them against direct biodiversity measures.