Unsupervised stroke clustering for developing common stroke set for percussion in Indian art music

Percussion instruments, such as the mridangam and table, play a vital role in Indian Art Music. While traditionally used as an accompaniment to maintain tempo and mark rhythmic progressions, these instruments are also featured in solo performances. Each stroke on these instruments, generated by a specific finger technique, produces distinct timbres, which are orally transmitted through mnemonic syllables-known as solkattu in mridangam and bol in tabla. However, a single stroke can be represented by different syllables across various schools and artists, creating a one-to-many mapping. This study addresses this issue by proposing an unsupervised, timbre-based clustering approach to develop a common stroke label set applicable across diverse notational systems. Such uniform transcription is crucial for rhythmic analysis and comparative studies within the field of Music Information Retrieval. Additionally, this work explores downstream tasks to validate the consistency and applicability of the derived stroke labels.