Data and Code for: An Intelligent Safety Companion: Real-Time Threat Detection Using Motion, Voice and Contextual AI

"This repository contains supporting data for 'An Intelligent Safety Companion: Real-Time Threat Detection Using Motion, Voice and Contextual AI.' It includes: (1) crime_dataset.csv — 50,500 georeferenced incident records (latitude, longitude, hour, crime type, severity) used to train and evaluate the KNN-based safe-routing model described in the paper; (2) Audio_Dataset3 — 88 short distress-phrase recordings in English, Hindi/Urdu, and Kannada, self-recorded by the authors and used for real-world validation of the wearable's end-to-end alert pipeline (not used in training or evaluating the CNN speech-emotion-recognition classifier. The SER classifier itself was trained and evaluated on the public RAVDESS, TESS, SAVEE, and CREMA-D corpora, which are not redistributed here due to their own licensing terms