Note: This bibliographic page is archived and will no longer be updated. For an up-to-date list of publications from the Music Technology Group see the Publications list .

Discovery of Percussion Patterns from Tabla Solo Recordings

Title Discovery of Percussion Patterns from Tabla Solo Recordings
Publication Type Master Thesis
Year of Publication 2015
Authors Gupta, S.
Abstract This thesis aims to explore the problem of percussion pattern discovery in Indian art music specific to Hindustani music. We propose approaches for automatic discovery of percussion patterns from audio recordings of tabla solos. We take advantage of the onomatopoeic syllables used in transmission of the repertoire and the technique. We provide a short introduction to the concept of rhythm in Hindustani music, explaining the essential components that constitute a rhythmic cycle. We review the relevant research for transcription and pattern search in music using computational methods and present a critique. A detailed description and the implementation steps for the proposed methods are presented. In our approach, the recordings of the Tabla solos are transcribed into a sequence of syllables using an HMM model of each syllable. We compile a set of most frequently occurring patterns in the dataset and use them as query patterns to be searched over the transcribed sequence of syllables. For the purpose of pattern search, we use an approximate string search method of matching Rough Longest Common Subsequence (RLCS) which, empirically, seems to be robust to the transcription errors and improves the recall rate and the f-measure significantly over the baseline. Further, we hypothesize to aid the RLCS approach by introducing syllabic similarity and non-linear costs for insertions and deletions in transcription. While the introduction of syllabic similarities gives a boost to the recall rate at the cost of precision, non-linear cost for insertions and deletions improves the precision at the cost of recall. The methods are evaluated over a sizeable dataset that has been compiled as a part of the CompMusic project. The obtained results provide a proof for the proposed concept. A detailed error analysis is performed and plausible reasons are discussed to explain the obtained results. The thesis concludes with a summary of the work, highlighting the main conclusions and the contributions made.