Certus Innovations Submission Pushes the Limits Further in Gunshot Audio ML
Certus Innovations is excited to announce the availability of the teams’ submission to DCASE 2026. DCASE, Detection and Classification of Acoustic Scenes and Events, is an area of research developing at a rapid pace to further environmental sound classification and detection.
The submission, “Exploring Feature Extraction Technique Parameters for Acoustic Gunshot Classification”, aims to revolutionize a key feature in developing gunshot audio machine learning (ML): audio feature extraction.
Audio feature extraction, the process of turning raw sound into measurements and data, makes it easier for a machine learning model to recognize patterns.
For the best results of gunshot audio ML, expertise is required in 3 fields: firearms, audio, and machine learning. Certus’s C3GD provides the largest and most detailed gunshot audio dataset, covering the “firearms” part of the equation, and this preprint, which has been submitted to DCASE 2026, aims to do the same for the audio piece of the puzzle.
In the paper, Gurny and Quinn benchmark three feature extraction techniques and demonstrate how choosing the correct technique can improve model accuracy by up to 20%. While no specific approach to feature extraction dominates universally, Gurny and Quinn validate why finding the best technique and parameters is an integral step in model development to find the best technique and parameters.
The preprint is available below:
Gurny, Sinclair, and Ryan Quinn. “Exploring Feature Extraction Technique Parameters for Acoustic Gunshot Classification.” arXiv: 2606.19568. Preprint, arXiv, June 17, 2026. https://doi.org/10.48550/arXiv.2606.19568.
Looking forward, while the field of digital signal processing has a plethora to offer, it is ineffectively supported by current tools. Certus Innovations is currently working on developing the final piece required in gunshot audio ML, pushing beyond what’s possible with the techniques currently available in ML literature.
