Abstract
Road traffic accidents remain a significant issue in Kazakhstan, with delayed emergency response times and a lack of affordable detection systems. Existing solutions, such as Advanced Traffic Management Systems (ATMS), are costly and unsuitable for older vehicles.This article presents the design and implementation of Saqtan, a mobile application aimed at improving road safety through real-time accident detection and automated alerting. The system leverages audio signal analysis and machine learning algorithms to recognize distinct patterns associated with car crashes. A custom dataset, supplemented with publicly available accident-related audio recordings, was used to train a Convolutional Neural Network (CNN) capable of classifying high-risk acoustic events. Key audio features, including Mel-Frequency Cepstral Coefficients (MFCC), spectral flux, and zero-crossing rate, were extracted to enhance detection accuracy. The application architecture supports automatic notifications to emergency contacts and services upon crash detection, incorporating a confirmation mechanism to mitigate false alerts. Experimental results demonstrate that the proposed system can effectively distinguish between accident-related and non-critical sounds. The findings suggest that mobile-based warning systems, supported by deep learning models, offer a promising direction for scalable and accessible road safety solutions.