Research on Deep Learning-Based Vocal Separation and Digital Synthesis in Choral Music
Ingegneria Sismica · 30.04.2026 · 42 (4) · 1-18
DOI: 10.65102/is2026643 ↗Kurz gefasst
Fachartikel in der Zeitschrift Ingegneria Sismica, online erschienen am 30.04.2026. Verfasst von Luanyu Zhao.
Abstract (Originalsprache)
With the advancement of computer performance, audio processing technology has also achieved significant development. This study explores the application of deep learning in audio separation and music synthesis, introducing deep recurrent neural networks into audio separation to construct a choral voice separation model. It further proposes a choral vocal synthesis system encompassing both training and synthesis, built upon deep neural networks. Experimental results demonstrate that the DNN-based choral voice separation algorithm effectively isolates vocals from accompaniment. The separated choral vocals achieve a significantly higher PESQ score of 3.32 compared to both the original track and accompaniment, outperforming comparison algorithms by 26.82% to 35.10%. The DNN model achieved musical metrics close to the original, with errors under 10%. Its generated choral music demonstrated lower TER values across all vocal parts compared to other algorithms, remaining within 5%, indicating excellent choral vocal synthesis performance. This method enables high-quality vocal separation and generation, showcasing its potential applications in music creation.
Zitieren
APA
Zhao, L. (2026). Research on Deep Learning-Based Vocal Separation and Digital Synthesis in Choral Music. Ingegneria Sismica, 42(4), 1-18. https://doi.org/10.65102/is2026643
Vancouver
Zhao L. Research on Deep Learning-Based Vocal Separation and Digital Synthesis in Choral Music. Ingegneria Sismica. 2026;42(4):1-18. doi:10.65102/is2026643
BibTeX
@article{Zhao2026,
author = {Zhao, Luanyu},
title = {Research on Deep Learning-Based Vocal Separation and Digital Synthesis in Choral Music},
journal = {Ingegneria Sismica},
year = {2026},
volume = {42},
number = {4},
pages = {1--18},
doi = {10.65102/is2026643}
}
EUVIMED