Music Generation and Classification from Chinese Musical Instruments Using Deep Learning
Joonho Kwak
Waseda International Christian School, Nakano-ku, Japan
Publication date: November 4, 2024
Waseda International Christian School, Nakano-ku, Japan
Publication date: November 4, 2024
DOI: http://doi.org/10.34614/JIYRC202418
ABSTRACT
Technological advancements have allowed people to use deep learning models to create and classify music with extremely high accuracy and precision. In this paper, a dataset was created from gathering audio files of traditional Chinese instruments from online websites. Then, feature extraction and data augmentation techniques were applied to enhance the dataset. Subsequently, various deep learning and machine learning algorithms such as LSTM, Sequential, kNN, and Logistic Regression were built to generate music and classify the instruments. The paper presents the feature extraction methods, data augmentation methods, machine learning models, and deep learning models; it also presents the results and possible explanations for the outcomes. Experimental results show that the deep learning models, particularly the Sequential Keras model, had an accuracy of around 99.21%. The LSTM model used to generate music could create an audio file with a similar timbre and sound to the instrument it was provided.
Technological advancements have allowed people to use deep learning models to create and classify music with extremely high accuracy and precision. In this paper, a dataset was created from gathering audio files of traditional Chinese instruments from online websites. Then, feature extraction and data augmentation techniques were applied to enhance the dataset. Subsequently, various deep learning and machine learning algorithms such as LSTM, Sequential, kNN, and Logistic Regression were built to generate music and classify the instruments. The paper presents the feature extraction methods, data augmentation methods, machine learning models, and deep learning models; it also presents the results and possible explanations for the outcomes. Experimental results show that the deep learning models, particularly the Sequential Keras model, had an accuracy of around 99.21%. The LSTM model used to generate music could create an audio file with a similar timbre and sound to the instrument it was provided.
