Automatic Generation of Music Captions

Zieleziński M, Łukasik E. Automatic Generation of Music Captions. In: Express Paper. Audio Engineering Society; 2025.pp. 343:1-. Available from: https://aes.org/publications/elibrary-page/?id=22894

@inproceedings{Zielezinski2025_22894,
  author = {Zieleziński, Mateusz and Łukasik, Ewa},
  title = {{Automatic Generation of Music Captions}},
  pages = {343:1-},
  year = {2025},
  month = may,
  publisher = {Audio Engineering Society},
  url = {https://aes.org/publications/elibrary-page/?id=22894}
}

TY  - CPAPER
TI  - Automatic Generation of Music Captions
AU  - Zieleziński, Mateusz
AU  - Łukasik, Ewa
SP  - 1
M1  - Paper 343
PY  - 2025
DA  - 2025/05/12
UR  - https://aes.org/publications/elibrary-page/?id=22894
PB  - Audio Engineering Society
LA  - en
AB  - This paper discusses the process of generating natural language music descriptions, known as captioning, using deep learning and Large Language Models. We aim to integrate a contemporary architecture called MERT, explicitly designed to create meaningful musical representation, to demonstrate that higher-quality embeddings can achieve competitive performance even with reduced data quantities. The captions used for training are from the state-of-the-art LP-MusicCaps dataset. A quantitative and subjective assessment of the quality of the created captions is performed, highlighting the differences between various models.
ER  -