Separation slang - Laboratory mice use low-frequency call repertoire during physical separation - VGG16-extracted call spectrogram features and UMAP plot

The discovery of a diverse repertoire of ultrasonic vocalizations (USVs) sparked interest in understanding their role in mouse social behavior. Social communication in mice is not just vocal, but multimodal and occurs mostly in close proximity. Aiming to unravel the impact direct physical interaction has on the vocal communication of same-sex mouse dyads, we separated mice through a divider preventing direct physical interaction, but allowing visual, olfactory and some tactile interaction through holes. Separated dyads emitted a distinct call repertoire consisting mainly of calls in or just above the human audible range (but not squeaks) as well as Noisy calls, and only to a lesser degree of USVs. This dataset contains sound recordings of single mouse calls, their spectrograms as images, spectro-temporal parameters of these calls, and VGG16-extracted spectrogram features. single mouse calls are stored as .wav files in tar archive 'wav_snippets'. Files are named according to the following scheme: mouse-pair-ID, mouse-pair-sex, serial call no., call type e.g. 276f_1_LFV-Flat.wav spectrograms of single mouse calls are stored as .png files in tar archive 'call_spectrograms'. Spectrogram image files are named after the corresponding .wav file, e.g. 276f_1_LFV-Flat.png. features_VGG16_flatten.npy contains VGG16-extracted call spectrogram features, obtained as described here: https://doi.org/10.1101/2025.09.26.678748 The spectro-temporal call parameters extracted by DeepSqueak as .xlsx files are stored in the .tar archive 'call_data'. .xlsx files are named: Prefixed_Data_mouse-pair-ID-and-sex, e.g. Prefixed_Data_276f.xlsx To generate spectrograms from .wav files, extract features using VGG16, and plot those using UMAP, please see https://github.com/tmb404/MouseCallClustering.git for JUPYTER Notebook.