Diving into Song Lyrics with Large Language Models: Unveiling Metadata Insights and Fueling Video Lyrics Generation

This thesis explores the application of Large Language Models (LLMs) in the domain of music content analysis and generation, particularly focusing on song lyrics. This includes the development of prompt engineering meth- ods for various tasks like theme identification, summarization, mood detec- tion, quote extraction and content moderation. Experimental validation is conducted to showcase the effectiveness of LLMs in analyzing and generat- ing lyrics-related content, while leveraging the latest research findings on the utilization of LLMs to assess the performance of other LLMs. Furthermore, the study explores the generation of lyric videos through iterative prompting techniques and introduces advanced prompting strategies for enhanced con- tent generation. In summary, this dissertation adds to the expanding body of research on the application of LLMs in creative content generation and high- lights their potential in revolutionizing the way music content is analyzed, interpreted and generated.