Markov-based genetic algorithm with [formula omitted]-greedy exploration for Indian classical music composition

Abstract

Recent technological advancements have made it possible to automate some of the intricate activities involved in music production. In this paper, a heuristic method is proposed for automatic composition of Indian Classical Music (ICM). Composing ICM necessitates consideration of raga constraints which distinguishes it from other genres of music. The raga-based music cannot be generated through random exploration of search space. More specifically, certain disordering in note combinations can violate the raga constraints. To address the aforesaid issue, the paper proposes a novel method by combining genetic algorithm with Markov chain model attuned to ICM sequences. Moreover, the proposed method uses ϵ-greedy strategy to balance the exploration and exploitation of musical search space. In order to preserve the musical characteristics, domain-specific genetic operators (i.e., crossover and mutation) are defined. The evaluation function based on music theory is used to direct the search to favorable tracks. The Kullback Liebler (KL) divergence metric is used to analyze the probability distribution of output and sample data sequence. The results indicate that by regulating the exploration rate, the method is capable of generating melodic sequences while maintaining the desired musical note distribution. Moreover, the method converges faster than the conventional genetic algorithm.