EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Pitch-Conditioned Cycle Gan With Harmonic Consistency for High-Fidelity Unpaired Musical Timbre

Vinay Krishna Polasa, K. Sravan Abhilash

Engineering, Technology & Applied Science Research · 11.08.2026 · 16 (4) · 37183-37189

DOI: 10.48084/etasr.18262 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Engineering, Technology & Applied Science Research, online erschienen am 11.08.2026. Verfasst von Vinay Krishna Polasa und 1 weiteren Autor:innen. Die Arbeit nennt 28 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The goal of musical timbre transfer is to adjust the instrumental identity of an audio signal to retain the musical content that is underlying, including pitch, melody, and expressive dynamics. Current methods of unpaired spectrogram, especially Cycle Generative Adversarial Network (CycleGAN) variants, are prone to pitch drift, harmonic artifacts, and loss of expressive information due to the lack of disentangling musical content and timbre. To overcome these weaknesses, this paper presents a Pitch-Conditioned CycleGAN with Harmonic Consistency (PC-CycleGAN-HC) system of unpaired musical timbre transfer. The proposed method presents two key components: (i) explicit pitch-conditioned translation, in which frame-level fundamental frequency is added as a supplementary conditioning channel to maintain melodic structure and expressive phrasing, and (ii) a harmonic consistency loss that limits the maintenance of F0-aligned harmonic energy distributions, which ensures spectral coherence and natural instrument resonance. Audio signals are modeled with 80-band log-Mel spectrograms, and the model is trained on adversarial, cycle-consistency, identity, loudness-aware, and harmonic consistency losses by using PatchGAN discriminators. Tests on the Spec2DataTest dataset of flute to piano and piano to flute conversion indicate that PC-CycleGAN-HC is significantly superior to a vanilla CycleGAN in mel-cepstral distortion, log-spectral distance, pitch RMSE, and multi-track listening MUSHRA. The MATLAB implementation can be reproduced in its entirety and used practically in research, music production, and education.

Zitieren

APA

Polasa, V. K., & Abhilash, K. S. (2026). Pitch-Conditioned Cycle Gan With Harmonic Consistency for High-Fidelity Unpaired Musical Timbre. Engineering, Technology & Applied Science Research, 16(4), 37183-37189. https://doi.org/10.48084/etasr.18262

Vancouver

Polasa VK, Abhilash KS. Pitch-Conditioned Cycle Gan With Harmonic Consistency for High-Fidelity Unpaired Musical Timbre. Engineering, Technology & Applied Science Research. 2026;16(4):37183-37189. doi:10.48084/etasr.18262

BibTeX
@article{Polasa2026,
  author = {Polasa, Vinay Krishna and Abhilash, K. Sravan},
  title = {Pitch-Conditioned Cycle Gan With Harmonic Consistency for High-Fidelity Unpaired Musical Timbre},
  journal = {Engineering, Technology & Applied Science Research},
  year = {2026},
  volume = {16},
  number = {4},
  pages = {37183--37189},
  doi = {10.48084/etasr.18262}
}
Kostenlos starten →