Pulkit Khandelwal

Social media platforms like Twitter, LinkedIn, and Facebook manage diverse content, including articles, messages, images, and videos. To ensure a safe and appropriate environment, these platforms utilize various methods to filter out content that violates their policies, focusing on dangerous, objectionable, and abusive material. Audio, a significant component of multimedia content, plays a critical role in identifying whether a video is spam or legitimate. Key audio elements like gunshots, explosions, screams, and hate speech, combined with corresponding video frames, can significantly improve the accuracy and precision of multimedia content moderation.

In this session, we will explore using deep neural networks, particularly Convolutional Neural Networks (CNNs), to extract relevant audio features from spectrograms and other audio representations. CNN-based models are adept at identifying local patterns and key audio elements essential for effective content moderation. We will investigate alternative machine learning models, including Support Vector Machines (SVMs) and Random Forests.