<p>Ensuring safety in public transportation is a critical challenge for various levels of government in Mexico, where theft continues to be a frequent and concerning threat. This paper presents a novel system, the Smart Alert System for Public Transport Robberies (SAS-PT), which leverages Deep Learning combined with Transfer Learning techniques to analyze and classify audio signals indicative of robbery attempts. Our approach adapts pre-trained neural network models, refining them to detect specific audio cues linked to theft scenarios. We evaluate the system’s effectiveness across various pre-trained architectures and examine factors such as ambient noise and audio fidelity to optimize performance in real-world conditions. Results from experimental trials demonstrate that the SAS-PT achieves high detection accuracy, offering a reliable mechanism for rapid emergency alerts and enhancing safety within public transport networks.</p>