Abstract
In the realm of audio signal processing, detecting impulsive urban sounds within continuous audio streams remains a complex endeavor. This paper presents an innovative solution via a hybrid Convolutional Neural Network - Recurrent Neural Network (CNN-RNN) framework. The CNN segments of the model target the extraction of spatial hierarchies from the audio, while the RNN components leverage their temporal memory capabilities to recognize and sequence sporadic urban auditory events. Drawing data from a rich collection of urban audio signals sourced from multiple global metropolises, the hybrid model was rigorously trained and evaluated. Findings underscore its superior detection accuracy in comparison to conventional methods, with a marked decrease in false positives. Moreover, the adaptability of the CNN-RNN architecture was evidenced by its consistent performance across diverse acoustic environments. This advancement in impulsive urban sound detection not only accentuates the efficacy of combined CNN-RNN systems but also has broader implications for urban planning, environmental assessments, and public safety initiatives.