Article: Image-based digital processing technology and system for music signals Journal: International Journal of Arts and Technology (IJART) 2026 Vol.16 No.3 pp.301 - 320 Abstract: Traditional music signal digital processing only focuses on the time-frequency characteristics of audio signals, ignoring the image features related to music signals, which leads to limitations in the overall understanding and representation of music signals. By comprehensively utilising the characteristics of audio and image signals, a more comprehensive and accurate music signal processing method was proposed to comprehensively and accurately understand music emotions. The music emotion data from MediaEval emotion in music, MagnaTagATune dataset, EmoReact dataset, and DEAM dataset were selected, pre-processed, and mapped to represent music emotions using valence and arousal. The long short-term memory (LSTM) - residual network (ResNet) model was constructed, with the LSTM module used to extract audio features and the ResNet module used to extract image features. In the fusion layer, the features extracted by the LSTM module and ResNet module were fused 1:1. A music recommendation system was constructed based on user historical preferences and recognition results of music emotions. The experimental results showed that the average accuracy of LSTM-ResNet model in music emotion classification on DEAM Dataset was as high as 98.5%. The combination of LSTM and ResNet can enhance the performance of music emotion classification and provide new methods for music recommendation tasks. Inderscience Publishers - linking academia, business and industry through research

Title: Image-based digital processing technology and system for music signals

Authors: Huan Li

Addresses: College of Music and Dance, Chongqing College of International Business and Economics, Hechuan District, 401520, Chongqing, China

Abstract: Traditional music signal digital processing only focuses on the time-frequency characteristics of audio signals, ignoring the image features related to music signals, which leads to limitations in the overall understanding and representation of music signals. By comprehensively utilising the characteristics of audio and image signals, a more comprehensive and accurate music signal processing method was proposed to comprehensively and accurately understand music emotions. The music emotion data from MediaEval emotion in music, MagnaTagATune dataset, EmoReact dataset, and DEAM dataset were selected, pre-processed, and mapped to represent music emotions using valence and arousal. The long short-term memory (LSTM) - residual network (ResNet) model was constructed, with the LSTM module used to extract audio features and the ResNet module used to extract image features. In the fusion layer, the features extracted by the LSTM module and ResNet module were fused 1:1. A music recommendation system was constructed based on user historical preferences and recognition results of music emotions. The experimental results showed that the average accuracy of LSTM-ResNet model in music emotion classification on DEAM Dataset was as high as 98.5%. The combination of LSTM and ResNet can enhance the performance of music emotion classification and provide new methods for music recommendation tasks.

Keywords: music signals; emotional classification; music recommendations; music images; long short-term memory; LSTM; residual network.

DOI: 10.1504/IJART.2026.155707

International Journal of Arts and Technology, 2026 Vol.16 No.3, pp.301 - 320

Received: 03 Apr 2025
Accepted: 21 Apr 2025
Published online: 11 Aug 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Image-based digital processing technology and system for music signals

Keep up-to-date