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

Area of Science:

• Ethnomusicology
• Music Information Retrieval
• Computational Musicology

Background:

• Indian Art Music heavily features percussion instruments like the mridangam and tabla.
• Rhythmic strokes are orally transmitted via mnemonic syllables (solkattu/bol), but inconsistencies exist across different schools and artists.
• This one-to-many mapping of strokes to syllables hinders comparative analysis and rhythmic studies.

Purpose of the Study:

• To develop a unified and consistent set of stroke labels for Indian percussion instruments.
• To address the ambiguity in oral transmission of rhythmic syllables.
• To facilitate rhythmic analysis and comparative studies in Music Information Retrieval.

Main Methods:

• An unsupervised, timbre-based clustering approach was employed.
• The method groups acoustically similar strokes regardless of their traditional mnemonic representation.
• Downstream tasks were used to validate the derived common stroke label set.

Main Results:

• A common stroke label set was successfully developed by clustering instrument timbres.
• The proposed method demonstrates consistency across diverse notational systems.
• Validation through downstream tasks confirmed the applicability of the unified labels.

Conclusions:

• The timbre-based clustering approach offers a robust solution for standardizing rhythmic stroke labels in Indian Art Music.
• This unified transcription system is essential for advancing Music Information Retrieval research.
• The study provides a foundation for more objective and comparative analyses of Indian percussion music.