Paper Title
MUSISYNTH: AN ARCHITECTURE THAT COMBINES NEURAL NETWORKS AND SYMBOLIC PROCESSING FOR ANALYZING INDIAN CLASSICAL MUSIC AND GENERATING RAGAS

Abstract
Creating computational models for Indian Classical Music poses a specific challenge because of the complicated rules associated with melodies in ICM and the unique, tradition-based musical framework, which can handle ICM in terms of analysis and synthesis. In terms of analysis, we have employed a double branch computational model where both acoustic and symbolic music theory characteristics are processed simultaneously. When it comes to synthesis, the framework uses a conditional LSTM-based model guided by style and raga. The novelty in our work lies in the use of an initialization method of the LSTM states, where the style and raga are directly embedded into the initial state of the LSTM network, making it easier for the network to produce melodies that are musically coherent right from the beginning. Keywords - Music information retrieval, Indian classical music, Deep learning, Hybrid neural-symbolic systems, Algorithmic Composition, Raga Recognition, Style Classification, Generative modelling, Sequence Learning, Computational Musicology.