| Title | : | An Approach to Transcription of Varnams in Carnatic Music using Hidden Markov Models |
| Speaker | : | Krishnaraj Sekhar P V (IITM) |
| Details | : | Wed, 18 Oct, 2017 3:00 PM @ A M Turing Hall |
| Abstract: | : | Carnatic music is an oral music tradition that has evolved over ten centuries or more into a rich musical genre. As music has been handed down from the teacher to the taught orally, there is hardly any annotation. Wherever annotation is available, the annotation is either ambiguous or incomplete. The objective of the proposed work is to provide rich and accurate transcription of Carnatic music. A varnam is always taught as part of standard lessons of Carnatic music as it is believed to embody the essence of the raga and tala in which it is set. The work proposes to develop a raga-specific transcription framework which uses tala information using varnam as the standard. Such a framework has the following applications: 1) Learning tool: Aspiring students can transcribe recordings of maestros and practice the various nuances as a self-learning and development exercise. 2) Documentation tool: Transcription of the notes and the gamaka serves as a tool to document the styles of various schools of music. Domain-specific information is used to build raga-specific HMMs. Rhythmic cycle(tala) information is used in the initialization of the HMMs, and also in the post-processing of the output. A dataset of varnams in eight ragas is used. The ground truth is obtained from standard notations of varnams. On an average 61.19% of accuracy, 52.42% of correctness and 45% of repetition-aligned accuracy is obtained across 8 ragas for transcription. |
