RideSafe: Detecting Sexual Harassment in Rideshares

摘要

Sexual harassment and abuse in rideshares is a growing problem. We propose a potential solution to this by using the voice recordings from the rideshare to detect cases of sexual harassment. Emotions such as fear, anger and disgust are most highly correlated to an individual being sexually harassed. Our solution aims to identify these emotions in a woman's voice as an indicator of sexual harassment. The Ryer-son Audio-Visual Database of Emotional Speech and Song dataset was used and offered voice recordings from male and female actors speaking sentences in different emotions. We extract the Mel-Frequency Cepstral Coefficient (MFCC) of the recordings in the dataset and run it through Machine Learning methods such as CNN (Convolutional Neural Network), SVM (Support Vector Machines) and LSTM (Long-Short Term Memory). We achieved an Fl-score of 95% with the CNN model on our dataset.