Detection of Hate Speech Detection in Videos Using Machine Learning

[Master's Project]

Objective:

To detect hate speech in videos based on the spoken content of the videos using machine learning.

Approach:

• Extract audio from video using FFmpeg API
• Convert audio to text transcript using Google Cloud Speech-to-Text API
• Extract features to get word counts, frequency of word counts, uni-grams and bi-grams
• Train Naive Bayes, Linear SVM, Random Forrest and RNN models
• Compute Accuracy, Precision score, Recall score, F1 score

Dataset:

• YouTube videos are searched and downloaded using a YouTube crawler
• Videos are searched using the 'search by keyword' function provided by YouTube Data API
• Searched videos are downloaded using Pytube library
• Each video is labeled as Normal or Hateful (Racist, Sexist)

Language Used:

• Python

Tools Used:

• YouTube Data API
• FFmpeg API
• Google Cloud Speech-to-Text API
• Google Storage Bucket
• Jupyter Notebook

Note: You can view .ipynb files using nbviewer - https://nbviewer.jupyter.org/