Application of Artificial Intelligence and Robotic Musicianship in South Indian Classical Music

This work utilizes Artificial Intelligence (AI) and Robotic Musicianship to address challenges specific to the field of Carnatic music - a music form popular in the southern part of India. For the past few decades, music AI and robotic researchers have focused on Western music with only a handful of attempts to address music from other parts of the world such as Carnatic music. This thesis aims to strengthen diversity and reduce bias in the field by researching and developing systems that understand and perform Carnatic music. Moreover, it bears promise to revolutionize productions in Carnatic music by providing software and hardware solutions to support gamakas - pitch-based embellishments which are at the core of this music form. It can also have a significant contribution to education, by introducing a violin-playing robot that can help students understand the nuances of gamaka playing in a repeated and systematic manner. A violin-playing robot can provide a much more accurate and expressive rendition of this music in comparison to software-based emulations. In this work, I designed and developed three tools for musicians and listeners of Carnatic music: 1. A 7-degrees-of-freedom violin-playing robot designed to play gamakas 2. A system to comprehend and generate gamakas 3. A system to use gamakas to automatically accompany a musician The robot uses pitch contour data to manipulate the left hand and amplitude contour data to make bow changes and modify dynamics. A system that comprehends and performs Carnatic music should understand how to interpret and create them. To achieve this, I developed GamakaNet - A novel Masked Latent Space Auto-Encoder model to synthesize gamakas for kalpitha swaras (Composed notes) in Carnatic music. A human violinist plays a major role in a typical Carnatic music concert by providing melodic accompaniment. An important characteristic of this music form is impromptu improvisations. One such section is raga alapana - a sequence of short characteristic phrases without rhythms that elevate the mood of a raga. I developed an AI based raga-agnostic algorithm to generate accompaniment for alapana. I additionally developed a clustering-based approach for sequencing characteristic melodic phrases for alapana in a given raga. This is applicable in the solo section of the accompanist during an alapana rendition. With the current Western-centric focus of AI music research, this thesis promises to enhance diversity and minimize bias, by exploring applications of AI and Robotics to the unique elements, datasets, and practices of Carnatic music.

Date

2024-08-20

Resource Type

Text

Resource Subtype

Dissertation

Full item page