The Swara-Net-5-Raaga Dataset

The dataset consists of 638 audio snippets covering five prominent ragas: Shankarabharanam, Revathi, Abheri, Brindavanasaranga, and Mayamalavagowla. All files were sourced from publicly available live performances and recordings on YouTube. To ensure high quality and validity, the collection was manually vetted by the author, a practitioner of Carnatic music, to verify each raga. Short snippets, approximately 15 to 30 seconds in length, were then manually extracted from the full recordings to capture the characteristic melodic phrases (raga-bhava) and to exclude non-representative sections such as percussion solos. A primary feature of this dataset is its timbral diversity. It was intentionally curated to contain a rich blend of vocal, instrumental (e.g., veena, violin, flute), and mixed performances. This resource provides a valuable, focused, and open-source collection for training and benchmarking machine learning models on this challenging and nuanced classification task