30.05.2022: House mice and many other animal species use ultrasonic vocalisations to communicate in various contexts including social and sexual interactions. These vocalisations are increasingly investigated in research on animal communication and for studying the genetic basis of autism and speech disorders. Researchers at the University of Veterinary Medicine Vienna have now developed a new and improved method for classifying ultrasonic vocalisations and are making the tool freely available to the scientific community.

As ultrasonic vocalizations (USVs) are above the range of human hearing (>20 kHz) and manual methods for analysing these vocalizations are extremely time consuming, several automated methods have been developed for USV detection and classification. In a study led by the Konrad Lorenz Institute of Ethology at the University of Veterinary Medicine Vienna, in collaboration with the Acoustics Research Institute, Austrian Academy of Science, researchers have now evaluated the advantages and disadvantages of the available analysis tools.

To investigate the performance of the available tools for analysing ultrasonic vocalisations, the researchers for the first time compared the available automatic detection methods and they used recordings from wild and laboratory mice. Also for the first time, the researchers analysed inter-observer reliability of manual methods as a gold standard to compare the performance of the automated methods.

BootSnap: new, free classification method

In addtion to evaluating detection tools, the researchers developed a new method for classifying USVs, which they call "BootSnap". The lead author, Reyhaneh Abbasi from the Konrad Lorenz Institute of Ethology at the University of Veterinary Medicine Vienna explains, “To improve USV classification, we developed an ensemble machine learning algorithm, which provides better performance than a single learning algorithm alone, and it allows us to successfully classify vocalisations into twelve types.” The researchers also tested the performance of their new method, and show that it provides greater generalisability than the current state-of-the-art tool. BootSnap is available free of charge for scientific use.

Reliable detection tool through use of large datasets

In contrast to previous detection methods, the researchers used a much larger sample of recordings and they used both wild house mice (Mus musculus) and laboratory mice for the detection of USVs prior to classification. Most USV detection tools that are currently employed use data from only one or a few strains of laboratory mice. “We found that our new methods provide significantly better overall performance compared to other detection and classification methods without the need for manual parameter tuning or special training,” says the study’s senior author, Dustin J. Penn from the Konrad Lorenz Institute of Ethology, University of Veterinary Medicine Vienna.

The article “Capturing the songs of mice with an improved detection and classification method for ultrasonic vocalizations (BootSnap)” by Reyhaneh Abbasi, Peter Balazs, Maria Adelaide Marconi, Doris Nicolakis, Sarah M. Zala and Dustin J. Penn was published in PLoS Computational Biology.

To the Scientific article