House mice emit complex ultrasonic vocalizations (USVs) during courtship and social interactions, although their functions are unclear. We hypothesize that USVs contain distinctive individual signatures that mediate individual recognition. If USVs contain individual signatures, it should be possible to develop a tool for assigning USVs to specific individuals when mice are interacting. Testing these ideas and conducting studies on USVs, in general, requires better methods for processing and analyzing USVs. The main goals of this project are to improve the processing of USV data and develop new and better tools for clustering and classifying USVs, analyzing their sequences, and recognizing mouse individuals and using these to determine the features of USVs that contain distinctive individual vocal signatures and test whether they influence individual vocal recognition.
Building upon our recently developed USVs classifier, BootSnap, we will employ signal processing algorithms, optimally adapted to mouse USVs, and develop deep learning approaches to improve USV classification accuracy. We will leverage these algorithms along with USV syntax analysis techniques to further enhance the accuracy of USV classification and learn more about USV sequences. Additionally, we will investigate the feasibility of recognizing individual mice based on their vocalizations through the development of deep learning algorithms. To ensure these methods’ effectiveness in various scenarios, these algorithms will undergo thorough analysis regarding their generalizability performance. The most successful approaches will be integrated into a new tool designed to streamline USV analysis and enable mouse individual recognition.
This interdisciplinary project utilizes signal processing and deep learning algorithms to enhance our understanding of mice hearing and advance research on USVs in general.
Duration of the project: 01.03.2023 - 28.02.2027