SqueakOut: Autoencoder-based segmentation of mouse ultrasonic vocalizations

• Paper @ bioRxiv
• USV segmentation dataset @ OSF

---

Inference-first repo. Supported workflow: load the pretrained checkpoint and segment a directory of spectrogram images.

Environment

Create or update the permanent squeakout micromamba environment:

micromamba env create -f environment.yml
micromamba env update -n squeakout -f environment.yml
micromamba run -n squeakout python -m ipykernel install --user --name squeakout --display-name "Python (squeakout)"

Usage

Notebook entrypoint: segmentation.ipynb

CLI entrypoint:

micromamba run -n squeakout python -m squeakout ./dataset/test

Legacy compatibility shims:

• from squeakout import SqueakOut_autoencoder
• from utils import create_montage

These are supported for old notebooks and scripts. Prefer the package API for new code.

Python API:

from pathlib import Path

from squeakout import load_model, resolve_device, segment_directory

device = resolve_device()
model = load_model(Path("./squeakout_weights.ckpt"), device=device)
results = segment_directory(
    Path("./dataset/test"),
    mask_root=Path("./outputs/segmentation"),
    montage_root=Path("./outputs/montages"),
    model=model,
    device=device,
)
print(f"saved {len(results)} masks")

Tests:

micromamba run -n squeakout pytest

---

BibTex citation

@article{Santana2024SqueakOut,
  title={SqueakOut: Autoencoder-based segmentation of mouse ultrasonic vocalizations},
  author={Gustavo M. Santana and Marcelo O. Dietrich},
  journal={bioRxiv},
  year={2024},
  doi={https://doi.org/10.1101/2024.04.19.590368},
  url={https://www.biorxiv.org/content/10.1101/2024.04.19.590368v1}
}