Abstract
This thesis explores the intersection of Artificial Intelligence (AI) and music production, focusing on the development of an AI-driven lyric generator capable of creatingcontextually appropriate and meaningful lyrics from symbolic music data. Motivated
by the recent advancements in generative AI models and their potential applications
in creative domains, this research aims to bridge the gap between AI capabilities and
lyric writing—a field traditionally dominated by human creativity. The methodology
integrates state-of-the-art AI technologies with music analysis techniques to classify
symbolic music data into genres and emotions, detect musical structures, and generate
lyrics using Large Language Models. A user-provided song topic serves as the foundation for lyric generation, ensuring relevance and thematic consistency. A large dataset
of MIDI files labeled with genre and emotion information is conducted containing genre
labels for 17,014 unique songs with 14 different genres, in addition to emotion labels
for 15,114 songs with the 5 basic emotions: anger, fear, joy, love, and sadness. These
datasets are used to train the AI models, which scored an accuracy of 0.43 for genre
classification and 0.65 for emotion classification, which is considered a good performance
in the context of music classification. This information is then used to generate lyrics
that are contextually aligned with the music’s genre and emotion. The results from
the implementation of this methodology show that, the AI-driven lyric generator can
produce lyrics that are coherent and contextually aligned with the music’s emotional
and structural characteristics. The evaluation of the generated lyrics using the BLEU
score and other lyric-specific metrics demonstrates that even though the AI-generated
lyrics may not match the quality of human-written lyrics, they exhibit a degree of relevance and coherence and can serve as a valuable tool for assisting human songwriters
and producers in the creative process. This study not only confirms the feasibility of
using AI for lyric generation but also contributes to the broader understanding of AI’s
potential in creative processes. The system developed offers a novel tool for artists and
producers, democratizing music production by enabling enhanced creative expression
through AI-assisted lyric writing. Future work will focus on refining the AI models and
expanding their applicability across diverse musical genres and languages.
Date of Award | 2024 |
---|---|
Original language | English (American) |
Supervisor | Christoph Schaffer (Supervisor) |