Using convolutional neural networks for acoustic-based emergency vehicle detection

Background: A siren is a special signal given by emergency vehicles such as fire trucks, police cars and ambulances to warn drivers or pedestrians on the road. However, drivers sometimes may not hear the siren due to the sound insulation of a modern car, the noise of city traffic, or their own inattention. This problem can lead to a delay in the provision of emergency services or even to traffic accidents. Aim: develop an acoustic method for detecting the presence of emergency vehicles on the road through the use of convolutional neural networks. Materials and Methods: The algorithm of work is based on the conversion of sound from the external environment into its spectrogram, for analysis by a convolutional neural network. An open dataset (Emergency Vehicle Siren Sounds) from sources available on Internet sites such as Google and Youtube, saved in the .wav audio format, was used as a dataset for siren sounds and city traffic. The code was developed on the Google.Colab platform using cloud storage. Results: The conducted experiments showed that the proposed method and model of the neural network make it possible to achieve an average efficiency of determining the type of sound with an accuracy of 93.3 % and a speed recognition of 0.00045 % of a second. Conclusion: The use of the developed technology for recognizing siren sounds in city noize will improve traffic safety and increase the chances of preventing a dangerous situation. Also, this system can be an additional assistant for hearing-impaired people while driving and everyday life for timely notification of the presence of emergency services nearby.