The first phase of the project involved data collection and processing, and the second phase involved audio analysis. The first dataset included test firings for approximately 20 firearms collected using multiple devices at multiple positions relative to the shooter. The second dataset included audio files extracted from YouTube, representing the type of "real world" data that might be encountered in casework. In data processing, several existing and newly written software tools were used with the two aforementioned databases to create a large set of individual processed files. Features of gunshot representation were then examined. Several approaches were considered for comparing the audio representations to complete each of the four proposed tasks. The analytical methods used to address each of the four main objectives are described. This involved gun-shot detection; shot-to-shot timings; the number of firearms present and gunshot assignment; and the prediction of firearm class, caliber, and make/model. Overall, this research demonstrated that useful information is contained within gunshot audio recordings; however, the information is difficult to extract from lower quality recordings and body cameras. This report concludes that the project was "fairly" successful in identifying gunshots, "extremely" successful in determining the number of different firearms present, and "moderately" successful in recognizing firearm class, caliber, and make/model. These results and the availability of the data collected may be a next step toward a fully automated gunshot analysis system. 8 tables and 5 figures
Downloads
Similar Publications
- Learning from Our Casework: The Forensic Anthropology Database for Assessing Methods Accuracy
- Transient Hypoxia Drives Soil Microbial Community Dynamics and Biogeochemistry During Human Decomposition
- Introducing the NIJ Forensic Intelligence Framework: Pillars and Guiding Principles for Successful Implementation