Introducing C3GD; The Largest, Most Detailed Gunshot Audio Dataset Ever Released for Open Science
Certus Innovations is excited to announce that the preprint of its IEEE Data Descriptions submission is live as of 6/16/26 on arXiv.
Certus Caliber Classification Gunshot Dataset (C3GD) is a publicly accessible data set developed for the analysis of firearm muzzle blast sounds. With metadata detailed beyond what is currently otherwise available, the dataset comprises a wide variety of firearms, calibers, cartridges, microphones, and microphone locations amounting to over 8000 field-collected data points from 28 firearms and across 16 calibers.
In part due to the costly nature of this type of research, much of the previous research on gunshot audio was collected from the internet. This largely underrepresentative historic data increased the risk of low-quality data and label noise, whereas C3GD aims to provide enough diversity to be able to generalize to more real-world applications. This is while also providing enough metadata for detailed academic analysis.
C3GD focuses on caliber-based classification rather than firearm model or broad categorical distinctions (e.g. pistol, rifle, shotgun), a design choice that enhances model generalization and aligns with recent advances in sensor technology, machine learning architectures, and digital signal processing techniques. Even within this comprehensive dataset All recordings underwent rigorous multi-stage quality control to ensure metadata accuracy, dataset reliability, and audio quality.
By making the dataset publicly available, Certus is showcasing their commitment to improving technology and reliability for defense innovation at large. Initiatives like C3GD aid the development of interoperable AI capabilities that can be shared, tested, and integrated across the defense community. Built with the same open philosophy that the TAK ecosystem prides itself on, C3GD reflects Certus’s mission in advancing open science and providing the high quality field data that would have been invaluable to the Certus team when this project began.
Finally, it is important to note that the C3GD dataset focuses primarily on the military force protection use case, so most mics are downrange of the shooter. For excellent public safety datasets with less detail but more off-axis recordings, see Kabealo et al and Lilien et al.
To access the full Certus Caliber Classification Gunshot Dataset publication, please see here on arXiv. Additionally, the full data set can be accessed from Zenodo. Thank you to Dr. Scott Friedman of SIFT for endorsing Certus to submit to the arXiv platform. Sift’s OpenIRIS/MIRA technology was used during TEAM-ML to track the provenance of data and catch possible issues before investing time and effort into training.
