Summary: Researchers have developed a new system that can automatically detect mouse song.
Source: University of Veterinary Medicine Vienna.
Mice, like some songbirds, produce complex vocalizations from an early age. Many of these calls occur in the ultrasonic range and are inaudible to humans, yet they form structured sequences that mice use to communicate in social and sexual contexts. Scientists study these ultrasonic vocalizations (USVs) both to understand natural mouse communication and as a model for neuropsychiatric disorders such as autism.
Traditionally, mouse USVs are recorded with ultrasonic microphones and analyzed manually—a process that is labor-intensive and subjective. Commercial software solutions exist, but their error rates and performance are not always well documented. Researchers at the Konrad Lorenz Institute of Ethology (Vetmeduni Vienna) and the Acoustics Research Institute (Austrian Academy of Sciences) have developed an automated detection tool called the Automatic Mouse Ultrasound Detector (A-MUD) to address these limitations.
The Automatic Mouse Ultrasound Detector (A-MUD)
Mouse “songs” vary in syllable type, duration, and temporal patterning. These differences allow mice to tailor vocal output to their social environment—for example, males often vocalize in response to the scent of a female. Because the structure and use of these vocalizations carry social meaning, researchers increasingly examine their composition and sequence, but progress has been slowed by time-consuming manual analysis and a lack of transparent benchmarks for existing automated tools.
The research team created an algorithm that runs within STx acoustic software (S_TOOLS-STx) and named it A-MUD. The tool was designed to process large datasets reliably and to match the quality of manual segmentation while operating much faster. In benchmark tests, A-MUD processed data several times faster than manual analysis and produced fewer detection errors than a commonly used commercial program. The authors made their tool freely available for scientific use and indicated ongoing work to develop improved versions.
Performance and error comparison
When evaluated against manually segmented files (used as a gold standard), A-MUD showed overall accuracy comparable to manual analysis. The algorithm produced slightly higher rates of false negatives—missed USVs—largely because it uses conservative thresholds to reduce background noise and minimize false positives. In contrast, the commercial software tested tended to produce more false positives, which can artificially alter perceived vocal patterns by adding spurious calls. False negatives are generally easier to handle during downstream analyses than false positives, since missing elements can sometimes be interpolated or accounted for, while false positives introduce misleading data points.
Focus on wild house mice and natural social contexts
Many previous studies have focused on laboratory strains of mice, but the vocal behavior of wild house mice in natural social settings remains relatively unexplored. Understanding how wild mice use USVs in real-world environments can reveal when and why different call types are produced and how vocal behavior adapts to ecological and social pressures. A-MUD was developed with this broader research agenda in mind, enabling scalable, consistent processing of recordings collected outside the laboratory. By making A-MUD freely available to other research groups, the authors aim to encourage standardized, comparable analyses across studies and habitats.
Source: Justin Dupuis — University of Veterinary Medicine Vienna.
Image source: Image credited to Vetmeduni Vienna.
Original research: “Automatic mouse ultrasound detector (A-MUD): A new tool for processing rodent vocalizations” by Sarah M. Zala, Doris Reitschmidt, Anton Noll, Peter Balazs, and Dustin J. Penn, published in PLOS ONE (published online July 20, 2017). The study describes development and validation of the A-MUD algorithm and compares its performance to manual segmentation and a commercial detection program.
University of Veterinary Medicine Vienna. Automatically Detecting Mouse Song. NeuroscienceNews. Published September 8, 2017. Accessed September 8, 2017.
Abstract
Automatic mouse ultrasound detector (A-MUD): A new tool for processing rodent vocalizations
House mice (Mus musculus) emit intricate ultrasonic vocalizations during social and sexual interactions that resemble bird song in their compositional variety and sequential structure. Manual segmentation of these recordings is slow and can introduce subjectivity, so the authors developed A-MUD, an automated detection script compatible with STx acoustic software. A-MUD increased processing efficiency—running roughly 4–12 times faster than manual segmentation depending on file size—and reduced overall detection errors relative to the evaluated commercial program. The algorithm yielded more true positives and fewer false positives and false negatives; most A-MUD errors were false negatives, reflecting a conservative threshold set to limit background noise. This work presents the first systematic comparison of automated USV detection error rates and offers an openly available tool to the scientific community, with further improvements planned.