A large annotated dataset of vocalizations by common marmosets

Area of Science:

• Primate communication
• Bioacoustics
• Animal behavior

Background:

• Non-human primates utilize complex vocalizations for intraspecies communication.
• Previous marmoset (Callithrix jacchus) vocalization studies faced limitations in sampling rates and data annotation.
• Advanced analyses, such as those using Deep Neural Networks (DNNs), require comprehensive and well-labeled datasets.

Purpose of the Study:

• To establish a large-scale, high-fidelity database of common marmoset vocalizations.
• To overcome previous limitations in data collection and annotation for marmoset communication research.
• To provide a valuable resource for future studies on primate vocal behavior and DNN applications.

Main Methods:

• Continuous recording of ~20 common marmosets over 40 months using a 96 kHz sampling rate.
• Compilation of over 800,000 vocalization files, totaling 253 hours of data.
• Annotation of approximately 215,000 calls with specific vocalization types and development of a trained classifier.

Main Results:

• A comprehensive dataset of common marmoset social vocalizations, covering their known repertoire.
• A substantial portion of the dataset (over 215,000 calls) is annotated, facilitating detailed analysis.
• Validation of the dataset through expert cross-examination of representative samples.

Conclusions:

• The created marmoset vocalization database is a significant advancement for primate communication research.
• The dataset and accompanying classifier enable sophisticated analyses, including those using Deep Neural Networks.
• This resource will accelerate investigations into the complexities of marmoset vocal behavior and communication systems.