Dietrich LabPhysiology of Behavior

Open-source tool

VocalMat

Analysis of mouse ultrasonic vocalizations using computer vision and machine learning — an open-source pipeline built in the lab.

Mice communicate with ultrasonic vocalizations (USVs) that carry rich information about social and developmental state. VocalMat turns raw audio into quantitative data by treating each vocalization as an image and applying computer vision and machine learning to detect and classify calls.

We use it to study mother–infant communication and how vocal behavior develops — a core method behind the lab’s developmental neuroscience projects. The tool is open source so other labs can analyze their own recordings with a consistent, reproducible pipeline.

Read the paper GitHubData (OSF)
Spectrogram (sonogram) of mouse ultrasonic vocalizations, colored by VocalMat's detected calls.

Spectrogram (sonogram) of USVs — VocalMat detects and classifies each call.

How it works

From audio to insight

Computer vision

Detects and segments ultrasonic vocalizations directly from spectrogram images.

Machine learning

Classifies call types automatically, reducing manual annotation.

Open source

Freely available for the research community to use and extend.

Reproducible

A consistent, quantitative pipeline for analyzing mouse USVs.

Also from the lab

SqueakOut

A companion tool for autoencoder-based segmentation of mouse ultrasonic vocalizations. Currently available as a preprint.

Read the SqueakOut preprint

Methods & tools publications