AUTHOR=Chen Lei , Xu Zhiyong , Zhao Zhao 

TITLE=Biotic sound SNR influence analysis on acoustic indices

JOURNAL=Frontiers in Remote Sensing

VOLUME=Volume 3 - 2022

YEAR=2023

URL=https://www.frontiersin.org/journals/remote-sensing/articles/10.3389/frsen.2022.1079223

DOI=10.3389/frsen.2022.1079223

ISSN=2673-6187

ABSTRACT=In recent years, passive acoustic monitoring (PAM) has become increasingly popular. Many acoustic indices (AIs) have been proposed for rapid biodiversity assessment (RBA), however, most AIs have been reported to be susceptible to abiotic sounds such as wind or rain noise when the biotic sound is masked, which greatly limits the application of these AIs. In this work, in order to take an insight into the influence mechanism of signal-to-noise ratio (SNR) on AIs, four most commonly used AIs, i.e., the bioacoustic index (BIO), the acoustic diversity index (ADI), the acoustic evenness index (AEI), and the acoustic complexity index (ACI), were investigated using controlled computational experiments with field recordings collected in a suburban park in Xuzhou, China, in which bird vocalizations were employed as typical biotic sounds. In the experiments, different SNR conditions were obtained by varying biotic sound intensities while keeping the background noise fixed. Experimental results showed that three indices (ADI, ACI, and BIO) decreased while the trend of AEI was in the opposite direction as SNR declined, which was owing to several factors summarized as follows. Firstly, as for ADI and AEI, the peak value in the spectrogram will no longer correspond to the biotic sounds of interest when SNR decreases to a certain extent, leading to erroneous results of the proportion of sound occurring in each frequency band. Secondly, in BIO calculation, the accumulation of the difference between the sound level within each frequency band and the minimum sound level will drop dramatically with reduced biotic sound intensities. Finally, the ACI calculation result relies on the ratio between total differences among all adjacent frames and the total sum of all frames within each temporal step and frequency bin in the spectrogram. With SNR decreasing, the biotic components contribution in both the total differences and the total sum presents a complex impact on the final ACI value. This work is helpful to more comprehensively interpret the values of the above AIs in a real-world environment and promote the applications of PAM in RBA.