Improvement of wail siren signal detection in noisy environment using Goertzel filter with sliding window

• Emergency vehicle detection in noisy environment from siren signal. • Uses Goertzel filter and sliding window for real-time siren detection. • Requires fewer computations and less runtime memory than existing detection methods. • Achieves ∼92.5% accuracy with 250 ms inference under noisy conditions (SNR > 10 dB). To enhance emergency response, it is essential for Intelligent Traffic Management Systems (ITMS) to detect siren signals in real time, even in the presence of background noise. This study proposes a pattern recognition framework that can be implemented as a microsystem within the space-and power-constrained environment of Internet of Things (IoT) devices without requiring feature extraction, as is typically required in traditional neural network-based methods. The proposed method detects wail siren signals in the 700–1700 Hz range by analysing eight sample periods during the 2-second rising phase across both the time and frequency domains. In each period, the frequency components ranged from 300 to 400 Hz above the initial frequency and were subdivided into 100 Hz bands for enhanced precision. A multistage detection system employs banks of Goertzel filters, with frequency ranges dynamically adjusted based on the signal patterns observed in the previous period. The system progresses through the sample periods if nearly half of the frequency bins within a filter range detect signals above a defined threshold; otherwise, it returns to its initial state. Compared to traditional Fourier-based spectrum analysis and advanced lightweight convolutional neural network (CNN) - based approaches, this method significantly reduces computational complexity, making it well suited for real-time deployment in resource-limited IoT environments. In this study, the method was validated using wail signals with varying rising patterns and combinations, including a 1:1 background noise ratio.