Entropy and Envelope Modulation Analysis for Detection of Fish Vocalizations in Complex Coastal Soundscapes

The development of automatic detection algorithms for fish vocalizations is essential for advancing passive acoustic monitoring and improving our understanding of marine soundscapes. We present an acoustic-feature-based detection framework that combines acoustic entropy and envelope modulation analysis to identify repetitive fish vocalizations in complex coastal environments. The framework was applied to passive acoustic recordings collected in the Cabo de San Antonio Marine Reserve (Spain), an area characterized by dynamic soundscapes and intense vessel traffic. Detection operates through two complementary configurations: an entropy-based mode that identifies temporally coherent low-entropy events, and a combined entropy–envelope modulation mode that validates detections using the relationship between the pulse repetition frequency (f0) and its second-order modulation component (2·f0). The combined configuration substantially reduced false-positive detections while simultaneously increasing true-positive detections and reducing false negatives. Event-level evaluation yielded a precision of 0.71, recall of 0.96, and F1-score of 0.82, compared with 0.52, 0.91, and 0.66, respectively, for the entropy-only configuration. Detected signals showed a marked nocturnal pattern with activity peaks around sunset. Their pulse period, peak frequency, and nocturnal occurrence were consistent with published descriptions of Sciaena umbra vocalizations, suggesting this species as a potential biological source while acknowledging that species-level attribution cannot be established from passive acoustic data alone. Overall, the proposed framework provides an interpretable feature-based approach for detecting structured fish vocalizations under the acoustically challenging conditions represented in this dataset.