TY  - JOUR
AU  - Sridhar, Rajeswari 
AU  - Subramanian, Manasa 
AU  - Lavanya, B. M. 
AU  - Malinidevi, B. 
AU  - Geetha, T. V. 
PY  - 2011
TI  - Latent Dirichlet Allocation Model for Raga Identification of Carnatic Music
JF  - Journal of Computer Science
VL  - 7
IS  - 11
DO  - 10.3844/jcssp.2011.1711.1716
UR  - https://thescipub.com/abstract/jcssp.2011.1711.1716
AB  - Problem statement: In this study the Raga of South Indian Carnatic music is determined by constructing a model. Raga is a pre-determined arrangement of notes, which is characterized by an Arohana and Avarohana, which is the ascending and descending arrangement of notes and Raga lakshana. Approach: In this study a Latent Dirichlet Allocation (LDA) model is constructed to identify the Raga of South Indian Carnatic music. LDA is an unsupervised statistical approach which is being used for document classification to determine the underlying topics in a given document. The construction of LDA is based on the assumption that the notes of a given music piece can be mapped to the words in a topic and the topics in a document can be mapped to the Raga. The identification of notes is very difficult due to the narrow range of frequency and the characteristics of Carnatic music. This inclined us in moving to a probabilistic approach for the identification of Raga. In this study we identify the notes of a given signal and using these notes and Raga lakshana, a probabilistic model in terms of LDA's parameters ∝ and θ are computed and constructed for every Raga by initially assuming a value which is constant for every Raga. This value of ∝ is cultured after determining θ for a given Raga. The θ of a given Raga is computed using the characteristic phrases which is a sequence of notes and is unique for a given Raga. During the Raga identification phase, the value of ∝ and θ are computed and is matched with the constructed LDA model to identify the given Raga. Results: Using this model, the Raga identification of Parent Ragas had a lower error rate than that of Child Raga. For parent Raga an average identification rate of 75% was achieved. Conclusion/Recommendations: The accuracy of the algorithm can be improved by using more features of Raga lakshana. After identifying the Raga, it can be used as features to be used by a Music Information Retrieval system.