Utilizing DeepSqueak for automatic detection and classification of mammalian vocalizations: a case study on primate...

Area of Science:

• Bioacoustics
• Animal Communication
• Computational Biology

Background:

• Manual analysis of animal vocalizations is subjective and time-consuming.
• Validated automated methods are needed for diverse species and non-specialists.
• DeepSqueak software was developed for rodent ultrasonic vocalizations.

Purpose of the Study:

• To test and validate DeepSqueak for automated analysis of primate ultrasonic vocalizations.
• To assess its ability to detect, cluster, and classify vocalizations.
• To determine generalizability to other taxa.

Main Methods:

• Trained DeepSqueak detectors on gray mouse lemur (Microcebus murinus) vocalizations.
• Applied filters to reduce noise and call fragments.
• Validated detector performance across call types, individuals, and recording quality.
• Tested detectors on a related species (Goodman's mouse lemur).
• Utilized supervised classification and unsupervised clustering.

Main Results:

• Achieved 91% correct detection of mouse lemur vocalizations after filtering.
• Successfully detected vocalizations in a closely related primate species.
• Supervised classifier achieved 93% accuracy in call type classification.
• Unsupervised clustering aligned with human-defined categories.

Conclusions:

• DeepSqueak is a viable tool for automated high-frequency/ultrasonic vocalization analysis in primates.
• The software generalizes beyond rodents to other taxa.
• A robust validation procedure for bioacoustics software was demonstrated.