@article{jajoria-etal-2026-audio,
    title = "Audio-Lyrics Alignment Dataset for {I}talian Arias",
    author = "Jajoria, Pushkar  and
      Graciotti, Arianna  and
      Casali, Giovanna  and
      Alabi, Jesujoba  and
      Delmonte, Rodolfo  and
      Pompilio, Angelo  and
      Tripodi, Rocco  and
      McDermott, James  and
      Klakow, Dietrich",
    editor = "Piperidis, Stelios  and
      Bel, N{\'u}ria  and
      van den Heuvel, Henk  and
      Ide, Nancy  and
      Krek, Simon  and
      Toral, Antonio",
    journal = "International Conference on Language Resources and Evaluation",
    volume = "main",
    month = may,
    year = "2026",
    address = "Palma de Mallorca, Spain",
    publisher = "ELRA Language Resource Association",
    url = "https://preview.aclanthology.org/ingest-lrec/2026.lrec-main.454/",
    pages = "5757--5766",
    abstract = "Aligning song lyrics with sung audio is challenging, especially for languages and music styles where annotated datasets are scarce. We address this gap by presenting the first dataset of Italian opera arias annotated with lyrics and time-stamps per word. The dataset comprises of 24 arias drawn from well-known operas of the 18th to 20th centuries with a total audio duration of nearly two hours. We benchmark both music alignment models and speech forced alignment models and show that existing methods face significant challenges on this dataset, with performance dropping by 45{\%} compared to other datasets. Multilingual and speech-based models exhibit relatively better performance on this dataset. We also evaluate few-shot fine-tuning of these models on the new dataset and find that, while it yields only marginal overall improvement, it produces localized gains on specific arias, suggesting that limited exposure helps the model adapt to some patterns but cannot fully overcome differences in language or musical style."
}