SANE : An index of anthropogenic noise levels for wildlife research in terrestrial ecosystems

Understanding the impact of noise pollution on wildlife is challenging due to the complexity of isolating the anthropogenic components from the soundscape. Soundscape studies usually employ acoustic indices that apply thresholds to isolate anthrophony from natural sounds. However, natural and anthropogenic sounds do not always conform to these thresholds, hindering both accuracy and comparability of the resulting indices. While significant progress has been made in the automated identification of acoustic events through artificial intelligence‐based classifiers, an effective method to quantify anthropogenic acoustic pressure is still lacking. We propose a novel index, the Selective Anthropogenic Noise Exposure (SANE), which leverages BirdNET deep neural network to isolate human‐related sounds from recordings. The index consists of the sum of the Median Amplitude Index of all human noise categories, thereby capturing both the intensity and cumulative impact of multiple disturbance events, while also enabling the decomposition of noise level across different categories of disturbance (e.g. traffic, human voices). We test SANE's performance in a real urban setting and through soundscape simulations and compare it with two frequency‐based indices and another artificial intelligence‐based acoustic index (CityAnthroNet). SANE was effective in describing human noises within areas characterized by different levels of urbanization, improving upon other indices' shortcomings. Additionally, SANE was robust at very fine temporal scales, precisely quantifying anthrophony levels for single recordings. Among the indices considered, SANE was the only one that remained insensitive to low‐frequency biophony, which confounded both frequency and artificial intelligence‐based metrics. By leveraging artificial intelligence‐based classifiers to detect multiple human‐made sound classes, the SANE index captures the intensity of anthropogenic noise. At the same time, natural sounds falling within the traditionally identified anthrophony range are ignored. Furthermore, SANE has the potential to assess the relative contribution of different noise types to overall acoustic pollution, opening new research avenues regarding the effects of acoustic pollution on wildlife.