Maris Basha
De-anonymize and polish README: add ICML 2026 paper info, arXiv DOI, cross-links to vocsim/* repos
4597073
---
dataset_info:
features:
- name: audio
dtype:
audio:
sampling_rate: 250000
- name: label
dtype: string
- name: subset
dtype: string
- name: index
dtype: int64
- name: speaker
dtype: string
- name: original_name
dtype: string
splits:
- name: train
num_bytes: 2083205598.525
num_examples: 31475
download_size: 1250386212
dataset_size: 2083205598.525
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
license: cc-by-4.0
tags:
- audio
- animal-vocalization
- ultrasonic-vocalization
- mouse
- bioacoustics
- classification
- benchmark
- vocsim
size_categories:
- 10K<n<100K
pretty_name: VocSim Mouse Strain Classification
---
# VocSim — Mouse Strain Classification
[![GitHub](https://img.shields.io/badge/GitHub-vocsim%2Fbenchmark-black?logo=github)](https://github.com/vocsim/benchmark)
[![Core dataset](https://img.shields.io/badge/%F0%9F%A4%97%20Core-vocsim%2Fpublic-blue)](https://huggingface.co/datasets/vocsim/public)
[![License: CC BY 4.0](https://img.shields.io/badge/License-CC%20BY%204.0-blue.svg)](https://creativecommons.org/licenses/by/4.0/)
A companion dataset for the **VocSim** benchmark that tests whether audio embeddings preserve **strain identity** in mouse ultrasonic vocalizations (USVs). It contains pre-segmented USV syllables from **C57BL/6J (C57)** and **DBA/2J (DBA)** mice, sampled at the native 250 kHz so high-frequency structure is preserved.
> Basha, M., Zai, A. T., Stoll, S., & Hahnloser, R. H. R. *VocSim: A Training-free Benchmark for Zero-shot Content Identity in Single-source Audio.* ICML 2026. [arXiv:2512.10120](https://doi.org/10.48550/arXiv.2512.10120)
## Task
Supervised binary classification: given an audio syllable (or features derived from it), predict the correct strain (`label` ∈ {`C57`, `DBA`}). In the paper we use this dataset to validate that VocSim-top embeddings transfer to a downstream bioacoustic task.
## Schema
```python
{
"audio": {"array": np.ndarray, "sampling_rate": 250000},
"subset": "mouse_strain",
"index": 101,
"speaker": "C57_file_001",
"label": "C57", # target: C57 or DBA
"original_name": "C57/C57_file_001.wav"
}
```
## Quick start
```python
from datasets import load_dataset
ds = load_dataset("vocsim/mouse-strain-classification-benchmark", split="train")
print(ds[0])
```
For end-to-end evaluation, use [github.com/vocsim/benchmark](https://github.com/vocsim/benchmark) — see `reproducibility/scripts/mouse_strain.py`.
## Source data
USV recordings and segmentation rely on MUPET (Van Segbroeck et al., 2017). Please cite both that work and the VocSim paper if you use this dataset.
## Citation
```bibtex
@inproceedings{basha2026vocsim,
title = {VocSim: A Training-free Benchmark for Zero-shot Content Identity in Single-source Audio},
author = {Basha, Maris and Zai, Anja T. and Stoll, Sabine and Hahnloser, Richard H. R.},
booktitle = {Proceedings of the 43rd International Conference on Machine Learning (ICML)},
year = {2026},
doi = {10.48550/arXiv.2512.10120}
}
@article{VanSegbroeck2017,
author = {Van Segbroeck, Maarten and Knoll, Aaron T. and Levitt, Patricia and Narayanan, Shrikanth},
title = {{MUPET}-Mouse Ultrasonic Profile ExTraction: A Signal Processing Tool for Rapid and Unsupervised Analysis of Ultrasonic Vocalizations},
journal = {Neuron},
volume = {94},
number = {3},
pages = {465--485.e5},
year = {2017},
doi = {10.1016/j.neuron.2017.04.018}
}
```