Despite their vulnerable status, manatee conservation efforts are hindered because of limited scientific data due to the data collection challenges. Bio-acoustic studies utilizing advanced computational analysis of long-term recordings offer a promising approach for determining the manatee’s presence and abundance. The following manuscript describes an artificial intelligence pipeline that effectively implements automatic manatee identification and counting. The first step is fine-tuning a pre-trained deep neuronal network using transfer learning to detect manatee sounds with a binary accuracy of 96%. In the second phase, the pipeline implements an unsupervised learning method to group acoustic features of the detected sounds with a clustering score of 72%. The article addresses the implications and the outcomes of testing this proof of concept under experimental conditions with passive acoustics monitoring data on the Caribbean coast of Costa Rica and Panama.