Thaat Classification Using Recurrent Neural Networks with Long Short-Term Memory and Support Vector Machine

Abstract

This research paper introduces a groundbreaking method for music classification, emphasizing thaats rather than the conventional raga-centric approach. A comprehensive range of audio features, including amplitude envelope, RMSE, STFT, spectral centroid, MFCC, spectral bandwidth, and zero-crossing rate, is meticulously used to capture thaats' distinct characteristics in Indian classical music. Importantly, the study predicts emotional responses linked with the identified thaats. The dataset encompasses a diverse collection of musical compositions, each representing unique thaats. Three classifier models - RNN-LSTM, SVM, and HMM - undergo thorough training and testing to evaluate their classification performance. Initial findings showcase promising accuracies, with the RNN-LSTM model achieving 85% and SVM performing at 78%. These results highlight the effectiveness of this innovative approach in accurately categorizing music based on thaats and predicting associated emotional responses, providing a fresh perspective on music analysis in Indian classical music.