Article: 3D-SCN: a high-fidelity stylised generation model for ethnic music Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.86 pp.34 - 61 Abstract: To address the challenges of structural coherence and stylistic encoding in ethnic music generation, this paper proposes a novel three-dimensional sequence convolutional network (3D-SCN). The model integrates the translational invariance of CNNs, the sequence modelling of bi-LSTMs, and the feature fusion of multi-head attention. Through multi-dimensional feature extraction and spatiotemporal modelling, it captures patterns in traditional Chinese pentatonic music. Experiments show the model outperforms baselines like music transformer, achieving 98.7% accuracy, a 0.941 F1-score, and 95.6 in expert-rated 'ethnicity'. It also demonstrates strong cross-cultural adaptability (91.3% cultural accuracy) and long-sequence generation capability (92.8% structural integrity). Despite computational complexity, this work provides an effective framework for digital music generation and heritage. Future work will focus on model lightweighting, dynamic memory integration multi-voice expansion. Inderscience Publishers - linking academia, business and industry through research

Title: 3D-SCN: a high-fidelity stylised generation model for ethnic music

Authors: Jingbo Liu; Wei Zhang

Addresses: School of Music, Instrumental Department, Hunan International Economics University, Changsha, 410205, China ' School of Music, Instrumental Department, Hunan International Economics University, Changsha, 410205, China

Abstract: To address the challenges of structural coherence and stylistic encoding in ethnic music generation, this paper proposes a novel three-dimensional sequence convolutional network (3D-SCN). The model integrates the translational invariance of CNNs, the sequence modelling of bi-LSTMs, and the feature fusion of multi-head attention. Through multi-dimensional feature extraction and spatiotemporal modelling, it captures patterns in traditional Chinese pentatonic music. Experiments show the model outperforms baselines like music transformer, achieving 98.7% accuracy, a 0.941 F1-score, and 95.6 in expert-rated 'ethnicity'. It also demonstrates strong cross-cultural adaptability (91.3% cultural accuracy) and long-sequence generation capability (92.8% structural integrity). Despite computational complexity, this work provides an effective framework for digital music generation and heritage. Future work will focus on model lightweighting, dynamic memory integration multi-voice expansion.

Keywords: three-dimensional sequence; convolutional network; folk music; stylisation.

DOI: 10.1504/IJICT.2026.155560

International Journal of Information and Communication Technology, 2026 Vol.27 No.86, pp.34 - 61

Received: 24 Feb 2026
Accepted: 21 Apr 2026
Published online: 05 Aug 2026 *

Title: 3D-SCN: a high-fidelity stylised generation model for ethnic music

Keep up-to-date