Development of a High-Quality Dolphin Whistle Signal Dataset and Clustering Analysis of Sousa Chinensis Calls

Dolphin communication signals facilitate social interactions and behaviors such as parenting and courtship. Understanding these signals is crucial for environmental protection and ecosystem monitoring including the estimation of the species abundance. This study develops a high-quality dolphin whistle signal dataset using passive acoustic monitoring techniques and employs the pYIN-based algorithm to extract pitch features from the whistle signals of the endangered Sousa chinensis species. Clustering analysis is subsequently performed to identify the primary call types and investigate the characteristics of various call types. The findings demonstrate the quality of the collected whistle signals as well as the efficacy of the proposed algorithm. Moreover, the clustering analysis successfully detects and categorizes different whistle types without human intervention. The clustering analysis achieves an average silhouette score of 0.42, indicating that the clusters effectively group similar whistle signals while distinguishing between different call types. This study lays the groundwork for future research on dolphin communication and supports conservation strategies by facilitating more efficient and effective monitoring of dolphin populations through PAM techniques.