A companion dataset for the VocSim benchmark that tests whether audio embeddings preserve individual identity in mouse ultrasonic vocalizations (USVs). It contains pre-segmented USV syllables from multiple individual mice (the speaker field), sampled at the native 250 kHz, derived from recordings by Van Segbroeck et al. (2017).
Basha, M., Zai, A. T., Stoll, S., & Hahnloser, R. H. R. VocSim: A Training-free Benchmark for Zero-shot Content… See the full description on the dataset page:
https://huggingface.co/datasets/vocsim/mouse-identity-classification-benchmark.
