This is a web demo for the paper "Disentangling Overlapping Sources: Improving Vocal and Violin Separation in Carnatic Music".
The paper has been submitted to the Workshop on Indian Music Analysis and Generative Applications (WIMAGA) at the 2025 International Conference on Acoustics, Speech, and Signal Processing (ICASSP).
Abstract: Separating the individual elements in a music mixture is an important tool in computational musicology, allowing for an improved analysis of music repertoires. In the context
of Carnatic Music, such task remains a challenge given the suboptimal generalization of existing music source separation systems to this style. Although multi-stem Carnatic recordings exist, these are mostly collected from the mixing console in live
performances, therefore the individual stems are not clean enough
to follow regular supervised training. Furthermore, there is a
strong melodic correlation between the singing voice and the
violin, as the later follows the melody sung by the singer during
the performance. Existing strategies to address such problem
struggle with source quality and only consider vocals. In this
work, we extend these efforts and achieve improved separation
while extending the separation targets to the violin, an important
source in the repertoire, and therefore cover the separation of
the most common melodic components in Carnatic Music. Code
and models are made available through compiam..
All audios in this demo are from the CMC dataset. These are not copyrighted, but please do not share this demo or the displayed audios,
as these are only for the purpose of this demo.
These samples are related to Table 1 in the paper.