Article: High quality controllable music generation technology based on improved transformer Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.40 pp.1 - 25 Abstract: With the rapid development of intelligent music generation, existing models still suffer from coarse emotional control, unnatural transitions, and weak style adaptability. To address these limitations, a controllable music generation model is proposed by integrating an emotion enhancement module with an improved transformer architecture, where flash attention is employed to improve long-sequence modelling efficiency. A dual-channel emotion input mechanism, combining discrete and continuous emotions, enables hierarchical and dynamic emotional guidance. Experimental results show an F1-score of 92.98%, a rhythm jump rate of 5.31%, and a note synchronisation offset below 1.64% on a public dataset, outperforming comparative methods. The mean squared error of emotional transitions is below 0.018, and the average subjective score reaches 9.21, indicating superior emotional expressiveness and structural coherence. The proposed framework effectively improves emotional controllability, temporal stability, and generation efficiency, providing a robust solution for controllable music generation and related applications. Inderscience Publishers - linking academia, business and industry through research

Title: High quality controllable music generation technology based on improved transformer

Authors: Chen Yuan

Addresses: The Conservatory of Music, Xinjiang Normal University, Urumqi, 830054, China

Abstract: With the rapid development of intelligent music generation, existing models still suffer from coarse emotional control, unnatural transitions, and weak style adaptability. To address these limitations, a controllable music generation model is proposed by integrating an emotion enhancement module with an improved transformer architecture, where flash attention is employed to improve long-sequence modelling efficiency. A dual-channel emotion input mechanism, combining discrete and continuous emotions, enables hierarchical and dynamic emotional guidance. Experimental results show an F1-score of 92.98%, a rhythm jump rate of 5.31%, and a note synchronisation offset below 1.64% on a public dataset, outperforming comparative methods. The mean squared error of emotional transitions is below 0.018, and the average subjective score reaches 9.21, indicating superior emotional expressiveness and structural coherence. The proposed framework effectively improves emotional controllability, temporal stability, and generation efficiency, providing a robust solution for controllable music generation and related applications.

Keywords: music generation; high-quality; transformer; discrete emotions; continuous emotions.

DOI: 10.1504/IJICT.2026.153318

International Journal of Information and Communication Technology, 2026 Vol.27 No.40, pp.1 - 25

Received: 17 Oct 2025
Accepted: 08 Jan 2026
Published online: 01 May 2026 *

Title: High quality controllable music generation technology based on improved transformer

Keep up-to-date