PCA-based denoising and automatic recognition of marine biological sounds to estimate Bio-voice Count Index for marine monitoring

Abstract

Passive acoustic monitoring faces methodological challenges when isolating biological signals from anthropogenically dominated marine soundscapes. To address this, we present two novel computational workflows: (1) a Principal Component Analysis (PCA)-driven noise reduction algorithm that selectively suppresses anthropogenic noise (e.g., vessel sounds) overlapping with biological frequency bands; and (2) an automatic Bio-voice Count Index (BCI) that quantifies target biological sounds through energy thresholding and adjustable frequency-weighting curve. We validated these methods using both synthetic soundscapes and 700 min of field recordings from coral reef ecosystems in Sanya, China. The PCA algorithm improved mean signal-to-noise ratios of field recordings by 5.3 dB (from 7.6 dB to 12.9 dB), effectively enhancing biological sound detectability. The BCI demonstrated strong correlations with biological metrics. When combined with the Acoustic Complexity Index, it improved the accuracy of fish abundance estimation compared to single-index approaches. Critically, our method reduces the analysis time by >90 % compared to manual methods. These tools provide ecologists with a reproducible framework for quantifying biodiversity in noisy environments, directly applicable to coral reef health monitoring and anthropogenic impact assessments.