Quantifying and mitigating recorder-induced variability in ecological acoustic indices

Abstract

AbstractDue to the complexity of soundscapes, Ecological Acoustic indices (EAI) are frequently used as metrics to summarize ecologically meaningful information from audio recordings. Recent technological advances have allowed the rapid development of many audio recording devices with significant hardware/firmware variations among brands, whose effects in calculating EAI have not yet be determined. In this work, we show how recordings of the same landscape with different devices effectively hinder reproducibility and produce contradictory results. To address these issues, we propose a preprocessing pipeline to reduce EAI variability resulting from different hardware without altering the target information in the audio. To this end, we tested eight EAI commonly used in soundscape analyses. We targeted three common cases of variability caused by recorder characteristics: sampling frequency, microphone gain variation, and frequency response. We quantified the difference in the probability density functions of each index among recorders according to the Kullback-Leibler divergence. As a result, our approach reduced up to 75% variations among recorders from different brands (AudioMoth and SongMeter) and identified the conditions in which these devices are comparable. In conclusion, we demonstrated that different devices effectively affect EAI and show how these variations can be mitigated.Graphical AbstractHighlightsAddressing recorder-induced biases in acoustic indices for improved reproducibility.Proposing an effective method to mitigate recorder-related biases.Evaluating pipeline proposed performance via acoustic index distribution analysis.