According to the Gun Violence Archive, there have been 296 mass shootings in the United States this year. Sadly, 2021 is set to be the deadliest year for gun violence in the United States in the past two decades.
Distinguishing between a dangerous audio event like a gunshot and a non-life threatening event like a bursting plastic bag can mean the difference between life and death. In addition, it can also determine whether or not to deploy public safety workers. Humans, as well as computers, often confuse the sounds of a bursting plastic bag with the real sounds of gunshots.
In recent years, there has been some hesitation in implementing some of the well-known available acoustic shot detection systems, as they can be expensive and often unreliable.
In an experimental study, researchers at the College of Engineering and Computer Study at Florida Atlantic University focused on the reliability of these detection systems with respect to the rate of false positives. A model’s ability to correctly discern sounds, even in the most subtle scenarios, will differentiate a well-trained model from a poorly efficient model.
With the daunting task of counting all sounds similar to a shooting sound, the researchers created a new data set consisting of audio recordings of plastic bag explosions collected in various environments and conditions, such as bag sizes. plastic and the distance of the recording. microphones. The recordings of the audio clips lasted from 400 to 600 milliseconds.
The researchers also developed a classification algorithm based on a convolutional neural network (CNN), as a reference, to illustrate the relevance of this data collection effort. The data was then used, along with a sound dataset of gunshots, to train a CNN-based classification model to differentiate life-threatening gunfire events from plastic bag explosion events. not life threatening.
Study results, published in the journal Sensors, show how fake gunshot sounds can easily confuse a gunshot detection system. Seventy-five percent of plastic bag noises were misclassified as gunshot noises. The deep learning-based classification model formed with a popular urban sound data set containing gunshot sounds could not distinguish the pop sounds of plastic bags from the sounds of gunshots. However, once the pop sounds of the plastic bags were injected into the training of the models, the researchers found that the CNN classification model worked well to distinguish the actual sounds of gunshots from the sounds of plastic bags.
âAs humans, we use additional sensory input and past experiences to identify sounds. Computers, on the other hand, are trained to decipher information that is often irrelevant or imperceptible to human ears, âsaid Hanqi Zhuang, Ph.D., senior author, professor and chair, Department of Electrical Engineering and Computer Science. , College of Engineering and Computer Science. âSimilar to how bats fly around objects when they transmit high-pitched sound waves that will bounce back to them at different time intervals, we used different environments to give the machine learning algorithm a better sense of differentiation of closely related sounds. “
For the study, gunshot-like sounds were recorded in locations where there was a probability that shots would be fired, which included a total of eight indoor and outdoor locations. The data collection process began by experimenting with various types of bags, with trash bags selected as the most suitable. Most of the audio clips were captured using six recording devices. To test to what extent a sound classification model could be confused by fake gunshots, the researchers trained the model without exposing it to the pops of plastic bags.
There were 374 sample shots originally used to train the model, which were obtained from the Urban Sounds database. The researchers used 10 classes from the database (gunshot, dog bark, kids playing, car horn, air conditioner, street music, siren, idle engine, jackhammer, and drill). After training, the model was then used to test its ability to reject plastic bag noises as real shooting sounds.
âThe high percentage of classification errors indicates that it is very difficult for a classification model to distinguish gunshot-like sounds, such as plastic bag noises, from actual sounds of gunshot. shot, âsaid Rajesh Baliram Singh, first author and doctorate holder. student in the electrical and computer engineering department of the FAU. “This justifies the process of developing a dataset that contains sounds similar to real gunshot sounds.”
In gunfire detection, having a database of a particular sound that can be confused with the sound of gunshots but which is rich in diversity can lead to a more efficient shot detection system. This concept motivated researchers to create a database of plastic bag explosion sounds. The higher the diversity of the same sound, the higher the probability that the machine learning algorithm correctly detects that specific sound.
“Improving the performance of a gunshot detection algorithm, in particular to reduce its false positive rate, will reduce the chances of treating harmless audio trigger events as perilous audio events involving firearms,” ââhe said. said Stella Batalama, Ph.D., Dean of the College. engineering and computer science. “This dataset developed by our researchers, along with the classification model they trained for gunshots and gunshot-like sounds, is an important step leading to far fewer false positives and the l ‘improving overall public safety by deploying critical personnel only when needed. “
The co-author of the study is Jeet Kiran Pawani, MS, who conducted the study at Georgia Tech.