Scholarlink

<p>This research paper explores the application of deep learning techniques to enhance various domains, with a specific focus on automatic music transcription (AMT). AMT involves converting audio signals into musical notation, and in this study, attention is directed towards the Thai xylophone, a traditional instrument in Thailand, played with hard mallets. The dataset for this research is a collection of 33 songs played on the Thai xylophone with hard mallets. The proposed AMT system leverages a combination of feature extraction methods (Mel-Spectrogram, MFCC, and CQT) and Convolutional Neural Network (CNN) architectures. This approach allows the CNN to learn informative representations from the song data set. Furthermore, the study compares CNN with Long Short-Term Memory (LSTM) and Deep Neural Network (DNN), demonstrating that CNN achieves superior performance, with F1-Scores of 87.37% for frame detection and 93.84% for onset detection.</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

Enhancing automatic music transcription of Thai xylophone music performed with hard mallets: a deep learning approach and comparative analysis

  • 期刊论文

  • 发表日期：

    2026年1月27日

查看全文
  • Apichai Huaysrijan,
  • Sunee Pongpinigpinyo

摘要

This research paper explores the application of deep learning techniques to enhance various domains, with a specific focus on automatic music transcription (AMT). AMT involves converting audio signals into musical notation, and in this study, attention is directed towards the Thai xylophone, a traditional instrument in Thailand, played with hard mallets. The dataset for this research is a collection of 33 songs played on the Thai xylophone with hard mallets. The proposed AMT system leverages a combination of feature extraction methods (Mel-Spectrogram, MFCC, and CQT) and Convolutional Neural Network (CNN) architectures. This approach allows the CNN to learn informative representations from the song data set. Furthermore, the study compares CNN with Long Short-Term Memory (LSTM) and Deep Neural Network (DNN), demonstrating that CNN achieves superior performance, with F1-Scores of 87.37% for frame detection and 93.84% for onset detection.