Raga Identification from Arohana and Avarohana Patterns using PNCC and Clusters

Abstract

This project seeks to create a method for automatically identifying Indian classical music ragas based on their Arohana (ascending scale) and Avarohana (descending scale) patterns. The suggested approach uses Power Normalized Cepstral Coefficients (PNCC) as feature vectors to capture the different properties of the ragas. The process involves extracting PNCC features from audio samples representing different ragas. These features are used to create clusters using the K-Means clustering method. The clustering method attempts to group comparable Arohana and Avarohana patterns, assisting in identifying ragas. The approach includes training a model capable of mapping retrieved PNCC features to specific ragas. The clustering gives the system better accuracy by recognizing patterns and similarities in the dataset. The trained model can then automatically determine the raga based on its Arohana and Avarohana patterns in the testing phase. The results show that using PNCC features and clustering techniques enhances the raga recognition system's robustness and accuracy, making it more accurate than the current computational methods. This project's applications include music recommendation systems, cataloging, and musicological research. It can be integrated with digital audio platforms to tag and categorize the music based on their ragas automatically.