A multi-firearm, multi-orientation audio dataset of gunshots

Area of Science:

• Acoustic Signal Processing
• Machine Learning for Security
• Audio Forensics

Background:

• Early firearm discharge detection is critical for situational awareness in civilian and military contexts.
• Existing gunshot audio datasets lack variety in recording device orientation and time synchronization, limiting their use for Direction of Arrival (DoA) analysis.
• Accurate firearm location and model determination are vital for effective response planning.

Purpose of the Study:

• To present a novel, time-synchronized audio dataset for gunshot detection and classification.
• To address the limitations of existing datasets by including multi-firearm and multi-orientation recordings.
• To facilitate research on acoustic-based firearm localization and identification.

Main Methods:

• Collected a multi-firearm, multi-orientation audio dataset in a semi-controlled outdoor firing range.
• Utilized multiple edge devices positioned around the firing range to capture audio.
• Ensured precise time synchronization across all recording devices.

Main Results:

• The dataset provides a comprehensive collection of gunshot audio with varied orientations and synchronized recordings.
• Enables robust analysis for Direction of Arrival (DoA) estimation.
• Supports advanced machine learning models for firearm detection and classification.

Conclusions:

• The presented dataset significantly enhances capabilities for real-world gunshot detection and localization.
• It overcomes key limitations of previous datasets, paving the way for improved acoustic-based security systems.
• This resource is valuable for research in audio forensics, defense, and public safety applications.