Bijay Khadka

Fusemachines AI Fellowship Capstone Project

Football Commentary Generation System

An end-to-end AI pipeline for automated football match commentary using computer vision, object tracking, LLM narration, and speech synthesis.

PythonRoboflowYOLOByteTrackGroq APIDeepGramComputer VisionNLP

Overview

The Football Commentary Generation System is an end-to-end AI pipeline built during the Fusemachines AI Fellowship. The system detects match entities, tracks movement, identifies football events, generates natural language commentary, and outputs speech.

What I Built

  • Used Roboflow’s pre-trained model with YOLO architecture for player, ball, and referee detection.
  • Integrated ByteTrack for multi-object tracking under occlusion and complex motion.
  • Designed rule-based event detection for passes, interceptions, blocks, and shots.
  • Generated play-by-play narration using Groq LLM with structured prompt engineering.
  • Integrated DeepGram Text-to-Speech for real-time audio commentary output.

Collaboration

This project was developed with project partner Sudip Shrestha as part of the Fusemachines AI Fellowship capstone.