VocalPy

VocalPy detects, classifies, and segments animal vocalizations from audio recordings. It ships a CLI backed by species-specific detection pipelines, bundled pretrained classifiers, and a neural segmentation stage (SqueakOut).

The pipeline runs in three stages: detection finds candidate vocalizations in the spectrogram using contrast- and morphology-based methods tuned per species; classification filters out noise and labels each remaining call by type using pretrained MobileNetV2 models; segmentation optionally runs SqueakOut to produce a pixel-level binary mask for each detected call. You can run detection alone, detection + classification, or all three.

Supported species: mouse, rat, guinea pig

VocalPy is inspired by VocalMat.

Start here