Advancing Acoustic Intelligence: DART Model Results

At Certus Innovations, our mission is to advance technologies that deliver measurable impact in the field. The DART (Distributed Acoustic Reasoner for TAK) project is a testament to that commitment—pushing the boundaries of what’s possible in sound-based machine learning.

Our DART models were trained on two unique datasets to evaluate accuracy and adaptability under realistic conditions:

• Dataset 1: 25 calibers → 92% top-1 accuracy for gunshot classification

• Dataset 2: 11 more common calibers → 95% top-1 accuracy for gunshot classification

These results underscore DART’s precision and performance across a wide range of acoustic environments and firearm types.

What sets DART apart is its real-world realism. Model training incorporated extensive data augmentation to simulate operational conditions—introducing artificial background noise, microphone distortions, filtering, and gain modulation. Both datasets also included suppressed and silenced gunshot audio where available, ensuring the system remains robust even in the most complex soundscapes.

Despite leveraging datasets that are larger and more challenging than those reported in existing literature, DART achieved equal or superior results, demonstrating not only technical capability but also Certus’ dedication to field-ready innovation.

DART represents more than a model—it’s a step forward in acoustic situational awareness, enabling faster, more informed decision-making for mission-critical environments and that is what Certus Innovations is all about. 

Learn more about DART, TEAM-ML, and our broader machine learning initiatives at certusinnovations.com.