E-mail: [email protected] Email : [email protected] Contribution : Matthieu MARECHAL et Rodelin EXAVIER
Link to app: https://mmatthieu1290-hate-speech-detection-from-appinproduction-dupsz5.streamlit.app/
GitHub to app: https://github.com/mmatthieu1290/hate_speech_detection_from_wav
Folder on google drive which contains the databases (the files are too big to be uploaded on GitHub): https://drive.google.com/drive/folders/1hsEim7Hmhzk_r9qbO9smL0goVBNMXC5g?usp=sharing
Project Description: This project involves developing an application that can identify the level of verbal aggression in certain conversations. This application could possibly be used in schools to detect places where acts of violence take place through insults. We trained three models (logistic regression, xgboost and neural network) using a balanced database containing 138723 texts.
Notebooks folder: folder that contains the notebooks used for pre-processing, supervised learning (logistic regression, xgboost), deep learning with a neural network.
Local_Deploy_With_Microphone folder: folder that contains all the files to locally develop an application that records communications by a microphone in a predefined time interval and gives an aggressiveness score using three models: logistic regression, xgboost and neural network.
APP_in_production_streamlit folder: folder that contains all the files to put in production on streamlit an application that reads a wav file containing conversations, divides it into sentence and gives for each sentence an aggressiveness score using three models: logistic regression, xgboost and neural network.