A Whisper and BETO-Bert Based Web Application for Classification of Emergency Calls

This study presents a comprehensive web-based solution for the classification of emergency calls using state-of-the-art deep learning models, aimed at enhancing the accuracy and speed of emergency response systems. The primary aim is to categorize emergencies more effectively, particularly those related to safety, enabling rapid identification and response in call centers. The methodology is structured into key phases: first, the transcription phase employs the Whisper model for precise speech-to-text conversion; then, data preprocessing ensures the removal of irrelevant characters, numerical data, and common phrases to refine the input. In the translation phase, careful attention is given to maintaining linguistic consistency between English and Spanish. During the segmentation phase, tokenization and attention masking are applied to enhance text structure. Finally, the classification phase utilizes the BETO model-a BERT variant fine-tuned for Spanish-to classify calls into specific emergency types, including 'Accident,' 'Crime,' and 'Violence.' The proposed solution achieved a classification accuracy of 95.7^%, supported by a learning rate optimization process.