Detection of Melodic Motifs in Hindustani Vocal Alap Recordings

Basic Information

Project Title: Detection of Melodic Motifs in Hindustani Vocal Alap Recordings
Name: Shreyas Nadkarni
Project: BTP
Semester(s): Spring 2022-23 (Sem 8 for me) (this was my BTP-II)
Guide: Prof. Preeti Rao

Abstract

Melodic motifs form an important defining characteristic of a raga in Hindustani classical music. Using computational techniques, it is possible to analyse the occurrence and nature of these motifs or phrases derived from the pakads (series of catch phrases) of the raga in a musical piece. In this project, the main problem is the extraction of occurrences of characteristic melodic motifs including note sequences and glides from alaps. 3 ragas have been chosen for the analysis with one characteristic motif in each. The different techniques of encoding the melody such as the complex auto-encoder (CAE) architecture and the pitch contour have been discussed. CAE-based experiments involving self-similarity matrix computation for repeated pattern discovery in a song have been conducted. For extraction of the occurrences of a given motif, a dynamic time warping (DTW) based subsequence search has been employed. A larger goal is to observe the audiovisual correspondence within musically defined units. This report aims to provide details of these techniques, details of the data used as well as the insights obtained through the analysis.

Any courses you completed relevant to the project

EE 679: Speech Processing, HS4114: Music Analysis through Computing, and other basic courses like DS 203, CS 419, EE 338

Describe your experience on the project

Motivation: Prof is also my DDP guide, so I would get a longer time horizon for the research. Reason for choosing this field is that I am highly interested in data science as well as in musicology, and this project is an overlap between my academic and extracurricular interests Type of work: Data handling, scripting using Pandas, NumPy, Librosa, etc for audio and time series data. A LOT of Python coding was involved. Workload: Heavy, we used to have weekly meetings and ma’am expects us to do the work that has been assigned Publication: Yes, we published a paper in ISMIR 2023 (International Society for Music Information Retrieval) which combined my work with the work done by a PhD student Extension work: Yes I will be working on the same project for my DDP (under CMINDS)

Describe your experience with the guide

The Prof was highly involved and she took a lot of efforts to ensure that what we (the PhD student and I) were doing were relevant and apt. We had one meeting a week during the semester (alongwith other updates on MS Teams), but towards the paper deadline this frequency gradually increased to one (or two) meetings a day. It was hectic before the paper deadline. She was very approachable Evaluation: report + presentation

Any advice for anyone considering a project under the same guide? Any other professors working in similar fields?

If you work under ma’am, your Python skills will reach a different level. You will become an expert at handling large amounts of data, parsing files, processing, storing, etc. You will get to implement ML and DL techniques as well, but the primary effort is managing the data and making it ready for further processing (sounds easy, but with the kind of intricate tasks involved, it becomes difficult sometimes). For people wishing to go into data science industry, this can be very rewarding. On the other hand, the effort expected from you will be significant, so it will not be a chill ride. You will be expected to attend meetings and give regular updates. In short, it will be a high-effort-high-reward kind of project. You will gain a lot of skills, and some research outputs in the form of a paper if it goes well.