Chord sense: Enhancing stylistic Chord Progression generation with fine-tuned transformers

Chord Progressions (CP) constitute a fundamental element within musical compositions. Skillful application of harmonies can captivate audiences through the colors and emotions they elicit. While existing research has predominantly focused on generating stylistically coherent CPs and accompaniments, relatively few studies have delved into the optimization of generating specific CPs of interest across diverse harmonic contexts. On this basis, this study aims to address this gap by fine-tuning a foundational CP model using datasets generated through three distinct strategies. Subsequently, the performances of the strategies are compared using both existing and novel evaluation metrics. According to the analysis, the results reveal that the model fine-tuned using the third strategy demonstrates proficiency in producing the target CPs across diverse contexts and modes of generation in a musically coherent manner. This approach opens up avenues for creative learning and sharing of stylistic chord progressions through exchanging customized fine-tuned chord models.