Iot Audio Analytics for Automated Safety System Using Deep Learning Techniques
Increasing crime rate is one of the major causes of concern worldwide today. Many of the robberies and crimes become unsolvable because of the delay in reporting the crime. As time progresses, there has been a rising need to make our streets smart. This paper aims at creating a device that reports a crime in real-time by capturing and analyzing the sounds in the environment. Sounds like gunshots, screams, breakage, etc. when detected, trigger the device which autonomously sends alerts to the emergency number. The system uses sensors to detect and capture the audio signals, which are then acted upon by machine learning classification algorithms to recognize them correctly. Once the received signal is classified, based on the urgency of the situation, the police can take immediate actions when reported. This system can be implemented in places like banks, jewelry shops, showrooms, etc. that are prone to thefts and require an efficient safety system. Previous findings on similar ideas have detailed the use of a microphone to record the audio signal, extract necessary features and classify the sound based on a machine learning model. This paper explores some of the major research done on CNN and RNN techniques used for audio classification followed by a comparison of the implementation of different models for audio classification using both CNN and RNN classifiers.
Keywords: Iot systems; surveillance system; deep learning; audio classification; neural networks; convolutional neural networks