Automatic Recognition of Siren Sound in Traffic

In this paper, the capabilities of a deep learning convolutional neural network (DL-CNN) to recognize the sounds of ambulance and fire engine sirens, compared to other sounds encountered in traffic, were analyzed. The topic is one of interest in the context in which semi-autonomous or fully autonomous vehicles will be on the streets in the near future. After automatically detecting a siren from an emergency vehicle, it could attract the attention of an inattentive driver who is listening to music or talking on the phone. If the car is autonomous, it must slow down and make room for the emergency vehicle to pass. The recognition rate of DL-CNN was about 90%, using MFCC cepstral coefficients extracted from a database with around 800 records.