Indian Classical Music is one of the greatest treasures of India. ‘The Ragas’ are at the core part of this music. The Ragas are melodic music compositions which follow certain rules and have a definite music structure . Classical Music lovers can often guess these ragas when a maestro sings a melodious tune that belongs to certain Raga.
Similarly , can a software or app be developed to find “Raga” of an audio clip ?
My project is to develop this “RagaFinder”
Here are the details :
Objective: To develop a system that would indentify Raga of an audio clip of Indian Classical Music played on Bansuri as monophonic instrument.
Methodology (in brief):
Part A: Transcription
i. Audio sample (potential format is mp3) collection of chosen ragas (5 in number).These Ragas are supposed to stand distinctly apart from each (for example, bhupali , yaman, bairavi)other in their attributes space . Attributes are to be identified.
ii. Filtering contents of accompanying instruments such as tabla and tanpura using software available of the shelf (for example, PRAAT).
iii. Detecting frequencies at given time instant
iv. Identifying the frequencies as sargam normalized with respect to “sa” (frequencies to be grouped in 12 classes namely: sa ,re, Re ,ga, Ga, ma, Ma, Pa ,dha , Dha ,ni ,Ni)
Part B: Identification of Raga
- Defining and identifying attributes of a Raga that uniquely defines it(such as jati, vadi, samvadi, pakad…)There could be fundamental and derived attributes(exploring, not finalised yet)
- Numerical quantification of the attributes of the Raga and creating table of attributes for every Raga. This would serve as the master attribute table for the concerned Raga.
- Creating table for the audio clip under test.
- Arriving at a method to find similarity score for table under the test and the master table.
- Deciding the threshold to qualify the closeness of the test audio to the master clip. This may involve a number of test samples and appropriate statistical method.
My grey areas/gaps:
- How to identify “sa” for the given audio clip? One clue for this could be extracting tanpura spectrum from the master audio clip as well as from the clip under test. The ‘sa’ frequency for the two may be different on account of two different tanpuras. However, the two would display same frequency ratios with base as their respective ‘sa’s.
- The output of the process of detecting the frequency contents (here, the output of PRAAT) could be a frequency v/s time plot, i. e., occurrence of various frequencies at each sampling interval. How to classify them in 12 discrete (frequency) levels pertaining to sargam?
- How to achieve steps 2 to 5 in part B?
- The Raga Guide – Suvarnalata Rao