VocalPy

VocalPy detects, classifies, and segments animal vocalizations from audio recordings. It is inspired by VocalMat.

Point it at a .wav file and it runs a full analysis pipeline: it chunks the audio for parallel processing, detects vocalizations using a spectrogram-based method tuned for the selected species, removes noise candidates with a pretrained classifier, assigns vocalization-type labels, and optionally produces binary segmentation masks for each detected call using SqueakOut. Results land in a CSV, per-vocal spectrogram images, and serialized recording objects — all in an output folder next to the audio file.

Supported species: mouse, rat, guinea pig

Installation

Requires Python 3.12 and Git LFS (bundled model checkpoints are stored via Git LFS — clone without it and the models will be missing).

With micromamba (recommended):

micromamba create -y -n vocalpy python=3.12
micromamba activate vocalpy
git clone https://github.com/gumadeiras/vocalpy.git
cd vocalpy
pip install --upgrade pip
pip install -r requirements-dev.txt

With venv:

python3.12 -m venv .venv
source .venv/bin/activate
git clone https://github.com/gumadeiras/vocalpy.git
cd vocalpy
pip install --upgrade pip
pip install -r requirements-dev.txt

Quick start

# Detect and classify mouse ultrasonic vocalizations (USVs)
vocalpy -p /path/to/recording.wav

# Use the rat pipeline instead
vocalpy -a rat -p /path/to/recording.wav

# Also run autoencoder-based segmentation — produces a binary mask per vocal saved under cnn_mask/
vocalpy -p /path/to/recording.wav --segmenter

# Save spectrogram-overlay images so you can manually verify detections
vocalpy -p /path/to/recording.wav -l

Outputs land in {audio_name}_outputs/ next to the audio file. The CSV is the quickest way to inspect results; the .vocalpy files let you reload the full recording object in Python for further analysis.

See the full documentation for CLI reference, output format, and operational checks.