Music and Audio Computing Lab

Singable Lyric Translation


Main Contributor: Haven Kim

Singable lyric translation (translating lyrics to be sung in different languages), unlike typical translation tasks, necessitates a delicate balance between singability and semantics. Research topics in the field of singable lyric translation include computational evaluation, dataset construction, comparative analysis, and automated translation.




I. Computational Evaluation

Our framework represents the first attempt to evaluate singable lyric translation, using the following criteria:

  1. Line Syllable Count Distance: Do the syllable counts align closely enough to maintain the integrity of melodies?
  2. Phoneme Repetition Similarity: Is the degree of phoneme repetition well preserved in the translated lyrics?
  3. Musical Structure Distance: Is the lyrical structure well maintained?
  4. Semantic Similarity: Are the original lyrics and the translated lyrics semantically related, though not identical?


II. Dataset Construction

The field of singable lyric translation is suffering from the lack of dataset. A singable lyric translation dataset that aligns the original and translated lyrics by line and section has been built, hoping to benefit singable lyric translation research, such as comparative analysis and automated translation.


III. Comparative Analysis

The characteristics of singable lyric translation vary depending on genre. Investigating how genre affects translation is a relatively unexplored topic.


IV. Automated Translation

A neural model that automatically translates lyric can be trained with different approaches. The choice of approaches influence the generated outcomes.




Related Publications

  • K-pop Lyric Translation: Dataset, Analysis, and Neural Modelling
    Haven Kim, Jongmin Jung, Dasaem Jeong, Juhan Nam
    Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING) [paper][dataset]
  • A Computational Evaluation Framework for Singable Lyric Translation
    Haven Kim, Kento Watanabe, Masataka Goto, Juhan Nam
    Proceedings of the 24st International Society for Music Information Retrieval Conference (ISMIR), 2023 [paper]