Scholarlink

<p>Indian Classical Music (ICM) is deeply expressive and culturally rich, but recognising and learning its intricate ragas requires expert training and is often inaccessible to learners. To address this gap, this study aims to develop an automated framework for Raga recognition using deep learning, facilitating scalable and AI-assisted music education. The proposed framework uses a data-driven approach, which incorporates preprocessing models such as pitch tracking, tonic normalisation, and note quantisation to identify musically significant attributes in the CompMusic Carnatic and Hindustani datasets. Long-Short memory (LSTM) and Gated Recurrent Unit (GRU) models are used to learn the dynamic and melodic patterns of Indian ragas. Classification robustness is also increased by the ensemble strategy (SRGM-Ensemble). The experimental findings show an accuracy of 93.33% on the CMD-10 dataset and 75.83% on the CMD-40 dataset compared to the traditional TDMS and PCD-based methods. Cross-dataset validation of Carnatic and Hindustani music has up to 78.2% accuracy, establishing the model’s generalisation capability. The analysis of qualitative errors by attention heatmap shows that the majority of misclassifications are between allied ragas that have tonal overlap. The results demonstrate the prospects of deep learning to reproduce the tricky melodic patterns of Indian classical music and bring about the creation of intelligent tutoring systems, music advisory systems, and computer archiving frameworks.</p>

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

Automated Recognition of Indian Classical Ragas Using Deep Learning

  • 期刊论文

  • 发表日期：

    2025年11月8日

查看全文
  • R. Bhagyalakshmi,
  • M. B. Anandaraju

摘要

Indian Classical Music (ICM) is deeply expressive and culturally rich, but recognising and learning its intricate ragas requires expert training and is often inaccessible to learners. To address this gap, this study aims to develop an automated framework for Raga recognition using deep learning, facilitating scalable and AI-assisted music education. The proposed framework uses a data-driven approach, which incorporates preprocessing models such as pitch tracking, tonic normalisation, and note quantisation to identify musically significant attributes in the CompMusic Carnatic and Hindustani datasets. Long-Short memory (LSTM) and Gated Recurrent Unit (GRU) models are used to learn the dynamic and melodic patterns of Indian ragas. Classification robustness is also increased by the ensemble strategy (SRGM-Ensemble). The experimental findings show an accuracy of 93.33% on the CMD-10 dataset and 75.83% on the CMD-40 dataset compared to the traditional TDMS and PCD-based methods. Cross-dataset validation of Carnatic and Hindustani music has up to 78.2% accuracy, establishing the model’s generalisation capability. The analysis of qualitative errors by attention heatmap shows that the majority of misclassifications are between allied ragas that have tonal overlap. The results demonstrate the prospects of deep learning to reproduce the tricky melodic patterns of Indian classical music and bring about the creation of intelligent tutoring systems, music advisory systems, and computer archiving frameworks.