A comparison of the Mouse Song Analyzer and DeepSqueak ultrasonic vocalization analysis systems in C57BL/6J, FVB.129, and FVB mice

Abstract

Vocal communication is an essential behavior in mammals and is relevant to human neurodevelopmental conditions. Mice produce communicative vocalizations, known as ultrasonic vocalizations (USVs), that can be recorded with various programs. The Mouse Song Analyzer is an automated analysis system, while DeepSqueak is a semi-automated system. We used data from C57BL/6J, FVB.129, and FVB mice to compare whether the DeepSqueak and Mouse Song Analyzer systems measure a similar total number, duration, and fundamental frequency of USVs. We found that the two systems detected a similar quantity of USVs for FVB.129 mice (r= .90, p< .001), but displayed lower correlations for C57BL/6J (r= .76, p< .001) and FVB mice (r= .60, p< .001). We also found that DeepSqueak detected significantly more USVs for C57BL/6J mice than the Mouse Song Analyzer. The two systems detected a similar duration of USVs for C57BL/6J (r= .82, p< .001), but lower correlations for FVB.129 (r= .13, p< .001) and FVB mice (r= .51, p< .01) were found, with DeepSqueak detecting significantly more USVs per each strain. We found lower than acceptable correlations for fundamental frequency in C57BL/6J (r= .54, p< .01), FVB.129 (r= .76, p< .001), and FVB mice (r= .07, p= .76), with the Mouse Song Analyzer detecting a significantly higher fundamental frequency for FVB.129 mice. These findings demonstrate that the strain of mouse used significantly affects the number, duration, and fundamental frequency of USVs that are detected between programs. Overall, we found that DeepSqueak is more accurate than the Mouse Song Analyzer.