Analyzing the Application of Generative Artificial Intelligence in Team Sports: a Case Study on Football

This master's thesis investigates the application of generative artificial intelligence (AI) in team sports, focusing on football as a case study. The primary goal is to design and evaluate a system for real-time automated commentary that integrates computer vision, natural language processing (NLP), and text-to-speech (TTS) technologies. Using a pre-built computer vision tool and a curated dataset from SoccerNet, this research focuses on comparing different approaches to text generation and voice synthesis. The study is structured into several stages: event detection leverages pre-trained computer vision models from SoccerNet for high-accuracy data extraction; text generation is explored using various large language models, while smart prioritization techniques are evaluated for their effectiveness in dynamic game scenarios. Finally, text-to-speech synthesis is conducted using state-of-the-art TTS models. Evaluation metrics include coherence and relevance of generated text, the prioritization of key moments, and the naturalness of synthesized speech. This research highlights the comparative strengths and limitations of the proposed methods, contributing to the development of scalable, efficient solutions for real-time AI commentary in sports.