Scholarlink

This research delves into the realm of Music Information Retrieval (MIR), employing advanced Deep Learning models to tackle the intricate task of raga identification within South Indian Classical Music, specifically Carnatic Music. In the expansive landscape of Indian classical music, ragas stand as the fundamental melodic frameworks, each encapsulating a distinct emotional and aesthetic essence within musical compositions. Bridging traditional musical knowledge with cutting-edge technology, our study involved the curation of a bespoke dataset comprising 600 audio clips, representing 20 Carnatic ragas. We meticulously extracted spectral and rhythmic features from this dataset, utilizing frequency analyses for spectral characteristics and capturing temporal intricacies for rhythmic nuances. The Deep Learning models, including the Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), CNN–LSTM hybrid network, Artificial Neural Network, and LSTM Autoencoder, were trained on these acoustic features. Significantly, our findings highlight the exceptional efficacy of the LSTM-Autoencoder model, boasting an impressive 94% accuracy. This research not only contributes to the evolving field of MIR but also emphasizes the potential of advanced Deep Learning techniques in unraveling and classifying the nuanced complexities of Indian classical music, particularly within the context of Carnatic ragas, thus advancing the understanding and preservation of this rich musical tradition.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Effective LSTM – Autoencoder Approach with Acoustic Features in Indian Classical Music Raga Recognition

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • R. R. Rajalaxmi,
  • P. P. Sudharsana,
  • R. Thangarajan

摘要

This research delves into the realm of Music Information Retrieval (MIR), employing advanced Deep Learning models to tackle the intricate task of raga identification within South Indian Classical Music, specifically Carnatic Music. In the expansive landscape of Indian classical music, ragas stand as the fundamental melodic frameworks, each encapsulating a distinct emotional and aesthetic essence within musical compositions. Bridging traditional musical knowledge with cutting-edge technology, our study involved the curation of a bespoke dataset comprising 600 audio clips, representing 20 Carnatic ragas. We meticulously extracted spectral and rhythmic features from this dataset, utilizing frequency analyses for spectral characteristics and capturing temporal intricacies for rhythmic nuances. The Deep Learning models, including the Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), CNN–LSTM hybrid network, Artificial Neural Network, and LSTM Autoencoder, were trained on these acoustic features. Significantly, our findings highlight the exceptional efficacy of the LSTM-Autoencoder model, boasting an impressive 94% accuracy. This research not only contributes to the evolving field of MIR but also emphasizes the potential of advanced Deep Learning techniques in unraveling and classifying the nuanced complexities of Indian classical music, particularly within the context of Carnatic ragas, thus advancing the understanding and preservation of this rich musical tradition.