Intelligent Music Score Generation Method Combines Short-Time Fourier Transform and Improved Convolutional Neural Network

Researchers from Xiamen University have proposed a new intelligent music score generation method that combines short-time Fourier transform and improved convolutional neural network. The method aims to address the limitations of existing music score generation methods, which are limited by scene limitations and the quality of their generated scores. The proposed method utilizes short-time Fourier transform to transform the time-frequency of music signals and then inputs the transformed information into an improved convolutional neural network model. This improved model enhances the accuracy and diversity of music score generation by introducing label enhancement strategy and internal convolution structure.

Key Takeaways:

• The proposed method utilizes short-time Fourier transform to transform the time-frequency of music signals and inputs the transformed information into an improved convolutional neural network model.
• The improved model introduces label enhancement strategy and internal convolution structure to enhance the accuracy and diversity of music score generation.
• The method has strong generalization ability and can effectively increase the quality of music score creation on various music datasets.
• The matching rate and complete rate of the generated score of the proposed method were 92% and 95%, respectively.
• The proposed method can improve the efficiency and quality of music score generation, and can help drum learners understand their own performance in time.
• The research has potential applications in music education, music therapy, and music composition.

Statistics:

• The proposed method achieved a matching rate of 92% and a complete rate of 95% in music score generation.
• The score generation time of the proposed method was only 1.05 seconds.
• The improved model enhances the accuracy and diversity of music score generation by introducing label enhancement strategy and internal convolution structure.
• The research has potential applications in various fields, including music education, music therapy, and music composition.

Sources:

• Intelligent generation method of drum music scores based on improved CNN and STFT
• Scientific Reports, 2025;15(1):35322
• Nature Portfolio, Heidelberger Platz 3, Berlin, 14197, Germany
• Yuting Ni, College of Arts, Xiamen University, Xiamen, 361005, People's Republic of China
• Scientific Reports - www.nature.com/srep/
• Nature Publishing Group - www.nature.com/