Multi-channel CNN-Based Rāga Recognition in Carnatic Music Using Sequential Aggregation Strategy.

A vital aspect of Indian classical music (ICM) is rāga, which serves as a melodic framework for compositions and improvisations for both traditions of classical music. In this work, we propose a CNN-based sliding window analysis on mel-spectrogram and modgdgram for rāga recognition in Carnatic music. The important contribution of the work is that the proposed method neither requires pitch extraction nor metadata for the estimation of rāga. CNN learns the representation of rāga from the patterns in the mel-spectrogram/modgdgram during training through a sliding-window analysis. We train and test the network on the sliced-mel-spectrogram/modgdgram of the original audio while the final inference is performed on the audio as a whole. The performance is evaluated on 15 rāgas from the CompMusic dataset. Two fusion paradigms, namely multi-channel and multi-modal frameworks, have been implemented to identify the potential of two feature representations. Out of the two approaches, multi-modal architecture reports a macro-F1 measure of 0.72, which is at par with the performance of the baseline sequence classification model. The performance is also compared with that of a transfer learning approach. [ABSTRACT FROM AUTHOR]