Seven-Class Sarali Varisai Audio and Feature Dataset for Tonic-Invariant Pattern Classification

<p dir="ltr">This dataset supports the research study titled “A Tonic-Invariant Framework for Sarali Varisai Pattern Classification Using Legendre-Based Melodic Contour Representation.” It was created for the automatic classification of seven Sarali Varisai exercise types in Carnatic music.</p><p dir="ltr">The dataset was collected by the authors and contains 276 vocal recordings distributed across seven classes, from Type 1 to Type 7.

The recordings were captured using a Sony ICD-PX470 digital voice recorder in 16-bit, 44.1 kHz WAV format. Each class represents a distinct combination of ascending and descending swara sequences, note-cluster organization, repetition, and pause placement.</p><p dir="ltr">The deposited files include anonymized audio recordings, class labels, anonymous participant identifiers, MFCC-based feature files, tonic-normalized Legendre polynomial features, participant-wise training and testing split information, model-performance results, and supporting documentation.