Bullying Activity Detection System

Client: Group Project

Role: Backend Developer

Tech Stack: Python, ONNX, React, Next.js, Express.js, MongoDB (Docker), MinIO (Docker), Tailwind CSS, Docker

Bullying Activity Detection System

Project Overview

This system aims to detect bullying activity in real-time by analyzing audio inputs to identify emotional distress and distinguish between aggressor and victim voices. Built as a group project for a 2025 course, it combines signal processing, emotion recognition, and speaker diarization to flag potential bullying incidents in school or online environments.

Problem Statement

Bullying often goes unnoticed until it’s too late. Traditional reporting relies on victims speaking up, which many hesitate to do. An automated system that passively monitors audio cues (e.g., tone, pitch, word choice) could provide early intervention opportunities while preserving privacy through on-device processing.

How It Is Done

The pipeline starts with audio capture via WebRTC, followed by preprocessing (noise reduction, segmentation). We use pre-trained models for emotion classification (anger, fear, sadness) and voice activity detection. Speaker diarization helps differentiate actors, while a rule-based engine flags high-risk interactions. The backend is built with Flask, and the frontend uses React for dashboard visualization. All sensitive processing is designed to run locally when possible.