Abstract: The traditional manual extraction method for dolphin whistles is very time-consuming, and depends on the operator’s experience, which makes it difficult to ensure the reliability and consistency of the extraction results. Although some automated extraction methods can significantly improve the processing efficiency, their robustness in the complex and variable ocean environments is still relatively poor. An efficient and robust automatic signal feature extraction algorithm is of great necessity for the biological research and passive acoustic monitoring of dolphins. This paper proposed an automatic micro-labeling extraction method for the frequency modulation signals of marine mammals. This method aims to reduce the human involvement, while ensuring comprehensive and accurate extraction of the characteristic parameters. The method was implemented in two steps: manual marking and automatic processing. First, an operator selected certain points on the time-frequency spectrum to provide precise guidance for the subsequent processing. The number of points depends on the complexity of the contour, and usually the starting, ending, and inflection points are sufficient. In the subsequent automated processing step, the full time-frequency contours were obtained, and the associated time and frequency parameters were saved. This methos was verified with real acoustic recordings, showing that it can accurately extract characteristic parameters: the mean relative error of frequency was 0.49%, with a maximum of 1.55%, while the mean relative error of duration was 0.08 %, with a maximum of 0.63%. With the proposed method, the workload can be reduced to one fourth of that of the manual method. Due to the high accuracy and stability, this method offers a reliable technical foundation for research on dolphin calls and their acoustic behavior.