Skip to main content

Machine Learning-based Acoustic Gunshot Classification

Primary Investigators:
Peter Volgyesi and Akos Ledeczi
Brief Description of Project:
An affordable and reliable acoustic gunshot alert system is needed especially for schools. Schools are noisy places and indoor acoustics is challenging due to reverberations. In addition, such a system must not have false positives for obvious reasons.These requirements make indoor acoustic gunshot classification a hard problem. Recent advances in deep learning may provide a solution. The project is aimed at developing a machine learning-based event classification methodology using a rich library of transient acoustic events.

Desired Qualifications: 
Deep Learning Special Topics Class
Familiarity with TensorFlow or other deep learning frameworks
Strong programming ability 

Nature of Supervision:
Work closely with PIs
A Brief Research Plan (period is for 10 weeks):
Week 1: Study literature of transient acoustic event classification
Week 2: Extend existing acoustic event library
Weeks 3-10: Develop, test and revise deep learning architectures for gunshot classification 

Number of Open Slots: 1
Contact Information:
Name: Peter Volgyesi
Department: Electrical Engineering and Computer Science