Wind‐robust sound event detection and denoising for bioacoustics. Issue 9 (12th July 2022)

• Record Type: Journal Article
• Authors: Juodakis, Julius; Marsland, Stephen
• Is Part Of: Methods in ecology and evolution. Volume 13:Issue 9(2022)
• Page Start: 2005
• Page End: 2017
• Publication Date: 2022-07-12
• Subjects: bioacoustics -- denoising -- passive acoustic monitoring -- sound event detection; Ecology -- Periodicals; Evolution -- Periodicals; 577
• DOI: 10.1111/2041-210X.13928
• Languages: English
• ISSNs: 2041-210X

Abstract

Abstract: Sound recordings are used in various ecological studies, including wildlife monitoring by acoustic surveys. Such surveys often require automatic detection of target sound events in the large amount of data produced. However, current processing methods, especially those relying on sound intensity for detection, are severely impacted by wind, which causes transient intensity peaks. The rapid dynamics of this noise invalidate standard noise estimators, and no satisfactory method for dealing with wind exists in bioacoustics, where simple training and generalization between conditions are important. We estimate the transient noise level by fitting short‐term spectrum models to a wavelet packet representation. This estimator is then combined with log‐spectral subtraction to stabilize the background level. The resulting adjusted wavelet series can be analysed by standard detectors. We use real data from long‐term acoustic monitoring to tune this workflow, demonstrate its denoising capabilities and test the improved detection in two population surveys of birds. The proposed short‐term estimator was more effective than standard (constant) noise estimates in both denoising and detection tasks. In the surveys, the noise‐robust workflow greatly reduced the number of false alarms. As a result, the survey efficiency (precision of the estimated call density) improved for both species. In contrast to existing methods, the proposed estimator can adjust for transient broadband noises without requiring additional hardware or extensive tuning to each species. It improved the detection workflow based on very little training data, making it particularly attractive for detection of rare species or general soundscape analysis.