This work proposes an repertoire-specific framework to address the task of vocal pitch extraction for Carnatic Music. The current baselines that are used to extract the vocal melody for Carnatic Music are heuristics-based and not optimized for this repertoire, therefore are prone to not generalize or scale properly, and may have several parameters to be configured.

The authors propose to extract the vocal melody using a state-of-the-art vocal pitch extraction model, however large amount of well-annotated data are required for that. The authors build a repertoire-tuned framework to automatically generate high-confidence annotations, which may be used to train and test data-driven models for this task.

In the paper, the authors provide evaluation of the trained models, showing improvement over the baseline approaches. The improved extracted melodies are also compared in the context of melodic pattern discovery, a very relevant task for the analysis of Carnatic Music, and the results suggest that the newly extracted pitch tracks allow for a more accurate discovery of vocal melodic patterns in the considered pieces.

Plaja-Roglans, G., Nuttall, T., Pearson, L., Serra, X., & Miron, M. (2023). Repertoire- Specific Vocal Pitch Data Generation for Improved Melodic Analysis of Carnatic Music. Transactions of the International Society for Music Information Retrieval, 6(1), 13–26. DOI: https://doi.org/10.5334/tismir.137