Implementation of neural network for outdoor sound surveillance

• Main Author: Soerjonoto, Albert
• Other Authors: Andy Khong W H
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/140348
• Institution: Nanyang Technological University

Microphones enable computers to receive audio signals as an input, and in turn, enable sound surveillance to be a domain of software engineering. One area where sound surveillance is critically useful is in the homeland security. Gunshot and gunfire can hardly be detected using machine vision, since the bullets would be too small and too fast. On the other hand, the loud sound that a gun makes allows it to stand out among other sound events. This project explores the fastest ways to process audio signal, to extract their features. Those features then are learned through the method of deep learning to be classified between one sound event and another. The feature that would be extracted would be in the form of log Mel spectrogram, and the neural network architecture that would be used is a modification of the two-stage sound event detection and localization to be used as a classifier for traffic and gunshot sounds.