CycleGAN-VC: Non-parallel Voice Conversion Using Cycle-Consistent Adversarial Networks
The DOI matches its Crossref record; 60 of the 61 references it cites were found in a registry or an index.
Abstract
We propose a non-parallel voice-conversion (VC) method that can learn a mapping from source to target speech without relying on parallel data. The proposed method is particularly noteworthy in that it is general purpose and high quality and works without any extra data, modules, or alignment procedure. Our method, called CycleGAN-VC, uses a cycle-consistent adversarial network (CycleGAN) with gated convolutional neural networks (CNNs) and an identity-mapping loss. A CycleGAN learns forward and inverse mappings simultaneously using adversarial and cycle-consistency losses. This makes it possible to find an optimal pseudo pair from non-parallel data. Furthermore, the adversarial loss can bring the converted speech close to the target one on the basis of indistinguishability without explicit density estimation. This allows to avoid over-smoothing caused by statistical averaging, which occurs in many conventional statistical model-based VC methods that represent data distribution explicitly. We configure a CycleGAN with gated CNNs and train it with an identity-mapping loss. This allows the mapping function to capture sequential and hierarchical structures while preserving linguistic information. We evaluated our method on a non-parallel VC task. An objective evaluation showed that the converted feature sequence was near natural in terms of global variance and modulation spectra, which are structural indicators highly correlated with subjective evaluation. A subjective evaluation showed that the quality of the converted speech was comparable to that obtained with a Gaussian mixture model-based parallel VC method even though CycleGAN-VC is trained under disadvantageous conditions (non-parallel and half the amount of data).
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“Recent research [59,60,61,62,63] in VC technologies has addressed long-standing challenges such as data scarcity, cross-lingual conversion, and real-time processing.”
From Audio Deepfake Detection: What Has Been Achieved and What Lies Ahead (2025) · cites this paper as [59]
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 1×
“Inspired by a previous study [49] for neural style transfer and super-resolution, we used the network that included downsampling, residual [50], and upsampling layers, as well as incorporating instance normalization [51].”
MethodAdam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946 · cited 1×
“We trained the network using the Adam optimizer [53] with a batch size of 1.”
MethodFully convolutional networks for semantic segmentation
Long, J., Shelhamer, E., & Darrell, T. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298965 · Cited by 38,834
Unpaired image-to-image translation using cycle-consistent adversarial networks
Zhu, J.-Y., Park, T., Isola, P., & Efros, A. A. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.244 · Cited by 23,602
Labaca-Castro, R. (2023) · Machine Learning under Malware Attack · doi:10.1007/978-3-658-40442-0_9 · Cited by 19,272
GAN(Generative Adversarial Nets)
(2017) · Journal of Japan Society for Fuzzy Theory and Intelligent Informatics · doi:10.3156/jsoft.29.5_177_2 · Cited by 19,270
(2010) · International Conference on Machine Learning · Cited by 13,995
Perceptual Losses for Real-Time Style Transfer and Super-Resolution
Johnson, J., Alahi, A., & Fei-Fei, L. (2016) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-46475-6_43 · Cited by 11,283 · cited 2×
“Inspired by a previous study [49] for neural style transfer and super-resolution, we used the network that included downsampling, residual [50], and upsampling layers, as well as incorporating instance normalization [51].”
MethodShi, W., Caballero, J., Huszar, F., Totz, J., Aitken, A. P., Bishop, R., Rueckert, D., & Wang, Z. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.207 · Cited by 7,570
“Least squares generative adversarial networks,”
Mao, X., Li, Q., Xie, H., Lau, R. Y. K., Wang, Z., & Smolley, S. P. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.304 · Cited by 5,437 · cited 1×
“To stabilize training, we used a least squares GAN [52], which replaces the negative log likelihood objective in Ladv by a least squares loss.”
Method
Research works printed but not linked: 1
- maas, 2013, Rectifier nonlinearities improve neural network acoustic models, Proc ICML Workshop
Cited by
Items 301–354 of 354
End-to-End Cross-Lingual Voice Conversion using CycleGAN for Low Resource Indian Languages
Khadse, P. S., Chandra, S., Mukherjee, S., Bharati, P., Pramanik, D., Aitawade, A., & Mandal, S. K. D. (2023) · International Conference on Speech Database and Assessments · doi:10.1109/o-cocosda60357.2023.10482914
“CycleGAN-VC [11] was developed for the purpose of monolingual voice conversion.”
MethodLiu, F., Wang, H., Li, X., & Zheng, C. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.4798860
NVCGAN: Leveraging Generative Adversarial Networks for Robust Voice Conversion
Zhang, G., Liu, J., Bi, W., Dongye, G., Zhang, L., Jing, M., & Yu, J. (2024) · Lecture notes in computer science · doi:10.1007/978-981-97-5666-7_28 · Cited by 1
VQStarGAN: An Unseen-to-Unseen, Zero-Shot, Non-Parallel, Unsupervised Voice Conversion
Yue, H.-Y., Lee, C.-H.-M., Mak, T.-Y., Lau, W.-Y., & Lam, K. (2024) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn60899.2024.10650185
“GAN-based approach, such as StarGANv2-VC [1], CycleGAN-VC [2], StarGAN-VC2 [3], focuses on non-parallel voice conversion using generator and discriminator to generate converted samples that are similar to the style of the target speaker.”
MethodHuang, Q. (2024) · International Conference on Automation and Computing (ICAC) · doi:10.1109/icac61394.2024.10718807
Ankita, & Shahnawazuddin, S. (2024) · Circuits Systems and Signal Processing · doi:10.1007/s00034-024-02896-8
Expressive Voice Conversion Using Perturbation and Autoencoder Loss
Chen, Z., Liu, Y., & Liu, G. (2024) · IEEE International Conference on Information, Communication and Networks (ICICN) · doi:10.1109/icicn62625.2024.10761574
Discrete Unit Based Masking For Improving Disentanglement in Voice Conversion
Lee, P. H., Ülgen, İ. R., & Şişman, B. (2024) · IEEE Spoken Language Technology Workshop (SLT) · doi:10.1109/slt61566.2024.10832297 · Cited by 2
“However, the introduction of advanced networks such as generative adversarial networks (GANs) [5,6] and variational autoencoders (VAEs) [7] opened up ways to train VC architectures without any parallel training data.”
BackgroundMusic2Fail: Transfer Music to Failed Recorder Style
Leong, C. I., Chung, I.-L., Chao, K.-F., Wang, J.-Y., Yang, Y.-H., & Jang, J.-S. R. (2024) · Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) · doi:10.1109/apsipaasc63619.2025.10849247
NeuralEVC: End-to-End High-Quality Expressive Voice Conversion
Cao, D., & Zhang, J. (2024) · International Conference on Electrical, Mechanical and Computer Engineering (ICEMCE) · doi:10.1109/icemce64157.2024.10862530
CSA-VC: Contrastive Learning With Selective Attention in Non-Parallel Voice Conversion
Prihasto, B., Lin, Y.-X., Le, P. T., Wang, J., & Wu, C. (2025) · IEEE Transactions on Audio Speech and Language Processing · doi:10.1109/taslpro.2025.3542288
“The latest development of GAN [23], [24], [25], [26], [27] succeeded in obtaining satisfactory results.”
Kishida, T., & Nakashika, T. (2025) · IEICE Transactions on Information and Systems · doi:10.1587/transinf.2024edp7206
USIAL-VC: A One-Shot Voice Conversion by U-Net-Based Encoder and Speaker Identity Adaptive Learning
Peng, Y., & Wang, Y. (2025) · Communications in computer and information science · doi:10.1007/978-981-96-4279-3_10
Generative Adversarial Networks Based Image Augmentation
Chaki, J. (2025) · SpringerBriefs in applied sciences and technology · doi:10.1007/978-981-96-5081-1_2
Deng, G. (2025) · Journal of Computer and Communications · doi:10.4236/jcc.2025.134020
Zero-shot non-parallel voice conversion using auto-encoder and adaptive instance normalization
Alaa, Y., Aref, M. M., & Alfonse, M. (2025) · Signal Image and Video Processing · doi:10.1007/s11760-025-04236-y
“One of the state-of-art VC models using GAN [3 ] is StarGAN-VC [4] which achieves the many-to-many VC task using a single GAN model by introducing a new loss called domain classification loss to the CycleGAN-VC [ 5 ] model.”
Audio-Visual Speech Synthesis Leveraging Capsule-Enhanced Generative Adversarial Network
Ghosh, S., & Jana, N. D. (2024) · Joint International Conference on Data Science & Management of Data (CODS-COMAD) · doi:10.1145/3703323.3703325
“Kameoka et al. proposed StarGAN-VC [18], which amalgamates the strengths derived from both the Cycle Variational Autoencoder (CVAE) [43] and CycleGAN [19] methodologies.”
Esposito, S., Sgandurra, D., Giampaolo Bella, A., & O’Keeffe, D. (2025) · Springer eBooks · doi:10.1007/978-3-031-88145-9_2
Puyalnithi, T., Prasanna Venkatesh, S., & Timothy Florian, L., V (2025) · International Conference on Cognitive Robotics and Intelligent Systems (ICC-ROBINS) · doi:10.1109/icc-robins64345.2025.11086327
“CycleGAN-VC [3] further advanced the field by removing the need for parallel data using cycle-consistent training, greatly enhancing practical usability.”
InfluentialFSD 2.0: Improved Fake Song Dataset for Unknown-domain Deepfake Detection
Cheng, H., Wang, H., Niu, L., Zhang, Y., Jiang, Z., Hou, X., Hao, X., Shen, Y., & Ye, L. (2025) · IEEE International Conference on Multimedia and Expo workshops · doi:10.1109/icmew68306.2025.11152048
“CycleGAN [11] enables cross-vocalist voice transformation without parallel corpora using adversarial learning, while StarGAN [12] supports multi-vocalist scenarios, improving flexibility and preserving content and style.”
Wav2Vec2-based Automatic Speech Recognition Systems for Filipino Children’s Speech
Santiago, F. J., Frilles, G. L., Ramirez, J. C., & Cajote, R. D. (2025) · International Symposium on Multimedia and Communication Technology (ISMAC) · doi:10.1109/ismac65024.2025.11175977
“However, children’s speech recognition (CSR) systems face unique obstacles like the acoustic variability of children’s speech [3], limited training data for low-resource languages like Filipino, and the disfluent speech patterns children exhibit.”
Tibetan Dialect Conversion Without Using Parallel Corpus
Hong, T., Liu, Y., Ma-bao, B., Duojie, R., Wang, X., Yu, Y., & Tashi, N. (2025) · International Conference on Control, Robotics and Intelligent System (CCRIS) · doi:10.1109/ccris65764.2025.11193048
AVENet: Disentangling Features by Approximating Average Features for Voice Conversion
Wang, W., Zhou, Y., Zhu, J., Ding, H., Xu, J., & Li, S. (2025) · IEEE International Conference on Multimedia and Expo (ICME) · doi:10.1109/icme59968.2025.11209204
“Some have sought to adapt methods from image style transfer tasks, utilizing Generative Adversarial Networks (GANs) [6], [7] for voice conversion tasks.”
Content-Style Disentangled Audio Style Transfer via Diffusion Model
Wang, Y., Lu, J., Chen, J., Zhang, X., Liang, Y., Du, Z., Shi, Q., & Huang, S. (2025) · IEEE International Conference on Multimedia and Expo (ICME) · doi:10.1109/icme59968.2025.11209789
“For instance, methods like CycleGAN-VC [9] and StarGAN-VC [10] have attempted to apply similar principles to audio, where the transfer process is modeled as a separate optimization of content loss and style loss.”
Deep Learning-Based Restoration of Voice-Converted Audio for Speech and Speaker Recognition
Liao, W., & Huang, D. (2025) · Lecture notes in computer science · doi:10.1007/978-981-95-4398-4_19
Kobayashi, S., Nose, T., & Ito, A. (2025) · Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) · doi:10.1109/apsipaasc65261.2025.11249275
“CycleGAN-VC [20] introduced an encoder-decoder structure with 1D CNNs that extend in the temporal direction to capture relationships in the feature dimension while preserving the temporal structure.”
The Hierarchical Disentangled Information framework enables the discovery of meaningful structures
Hasanabadi, M. R., & Gharavian, D. (2025) · Scientific Reports · doi:10.1038/s41598-025-25815-6
Non-autoregressive Any to Many Voice Conversion
Zhou, Y., Peng, L., & Chen, Q. H. (2025) · International Conference on Computer Science and Network Technology (ICCSNT) · doi:10.1109/iccsnt67644.2025.11291353
“In the encoder-decoder model based on posterior voice probability, the decoder part uses a structure similar to the autoregressive recurrent neural network decoder in Tacotron 2 [8] or an autoregressive structure based on Transformer [9].”
Adversarial Spectral Transfer: Cycle-Consistent Learning for standard-free NIR Data Conversion
Lu, X., Wang, D., Zong, G., Li, C., Feng, W., Feng, S., Zhao, L., & Shu, L. (2025) · IEEE International Conference on Industrial Informatics (INDIN) · doi:10.1109/indin64977.2025.11279651
“Its effectiveness has been validated in multiple sequence modeling tasks, such as language modeling [16] and speech modeling [17].”
Long, P., & Guo, X. (2026) · Springer eBooks · doi:10.1007/978-981-96-9404-4_12
Exploring generative adversarial networks for controllable musical audio synthesis
Nistal, J. (2022) · theses.fr (ABES) · doi:10.70675/834b6be0z22afz42cez8589z4190efea7654 · Cited by 2
A neural voice transformation framework for modification of pitch and intensity
Bous, F. (2023) · theses.fr (ABES) · doi:10.70675/5a1b2884z903dz4b97z8073z7614dbb58847 · Cited by 2
Hierarchical temporal learning for multi-instrument and orchestral audio synthesis
Caillon, A. (2023) · theses.fr (ABES) · doi:10.70675/749064bdzd893z49b3za0fczb85cacee55d0
Dataset Generation for Studying Deepfake Content (or Not) with AI Techniques
Falcón-López, S. A., Tobarra, L., Robles-Gómez, A., & Vargas, R. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-16995-2_11 · Cited by 1
CLARIS: Clear and Intelligible Speech from Whispered and Dysarthric Voices
Shah, N., Sonkar, Y., Subhash Karande, S., & Gandhi, V. (2026) · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems · doi:10.1145/3772318.3791734
“Many-to-one converters such as CycleGAN-VC [37, 38], parallel-data systems such as Parrotron [5], and adversarial approaches [24] have also been explored, but they often suffer from overfitting, data requirements, or training instability.”
Real Time Voice Cloning Using Enhanced TTS and Vocoder
Borgaonkar, D., Hemanth, M., Abhishek, P., Chandra, K. R., & Srijith, P. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-18349-1_24
MeanVoiceFlow: One-Step Nonparallel Voice Conversion with Mean Flows
Kaneko, T., Kameoka, H., Tanaka, K., & Kondo, Y. (2026) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp55912.2026.11460769
Self-citationCyclegan++: enhancing performance through parameter sharing and domain alignment
Mahmoudabadi, M., & Rezghi, M. (2026) · Multimedia Tools and Applications · doi:10.1007/s11042-026-21645-4
Instrument-to-Instrument Timbre Conversion Using MaskCycleGAN
Lai, W.-H., & Hsieh, C.-W. (2026) · International Conference on Information Networking · doi:10.1109/icoin68469.2026.11480919
“Building upon this concept, CycleGAN [17] was later proposed, and in 2018, Kaneko et al. introduced CycleGAN-VC [10], followed by the improved CycleGAN-VC2 in 2019 [11].”
PRL-DAS: Robust Heliox Speech Recognition for Unaligned Low-Resource Data
Chen, Y., Zhang, G., Wen, W., & Zhang, S. (2026) · Big Data and Cognitive Computing · doi:10.3390/bdcc10050157
Lwin, H. Y., Kumwilaisak, W., Hansakunbuntheung, C., & Thatphithakkul, N. (2026) · IEEE Open Journal of the Computer Society · doi:10.1109/ojcs.2026.3696985
“CycleGAN-VC variants [13], [14], [15] enable non-parallel voice conversion via cycle consistency.”
Vowel Space-Guided DDSP for Severity-Preserved Healthy-to-Pathological Voice Conversion
Ko, T.-H., Chu, M.-C., & Wu, C. (2026) · IEEE Transactions on Audio Speech and Language Processing · doi:10.1109/taslpro.2026.3703186
“Generative models typically require substantial training data to ensure stability and generalization [13], [14].”
Meenakshi, K., Varshney, H., Gupta, A., Khanna, T., Bhargava, D., & Rana, A. (2026) · International Conference on Sustainable and Futuristic Technologies (ICSFT) · doi:10.1109/icsft67370.2026.11566554
Machine Learning Methods for Audio Signal
Xu, Q., Zhu, B., Su, Y., Sun, Y., Wen, H., & Xu, K. (2026) · Springer eBooks · doi:10.1007/978-981-95-0836-5_3
López, S. A. F. (2026) · UNED repository
PrivSpeech: A Declarative Framework for Privacy-Preserving Speech Analytics
Zhong, R., & Chen, S. S. (2026) · Global Information Infrastructure and Networking Symposium · doi:10.1109/giis69881.2026.11585740
Generalization Analysis of YAMNet-DNN Architectures in Deepfake Audio Classification
Diash, H. D., Prasetya, D. A., Pratama, A. R., & Fahrudin, T. M. (2026) · International Journal of Advances in Data and Information Systems · doi:10.59395/ijadis.v7i2.1567
NeutralSpeech: Two-Stage Emotion Suppression for Enhanced ASR Performance
Lakkad, J. S., Tiwari, S. R., Savaliya, L. J., & Patel, P. J. (2026) · International Conference on Speech, Multimodal and Advanced Communication Systems (ICSMACS) · doi:10.1109/icsmacs68968.2026.11628528
“CycleGAN-VC [11] further evolved non-parallel VC by incorporating adversarial loss and cycle-consistency in learning mappings between different speaker or style domains, hence removing the need for paired data and enabling broad generalization.”
Dibbo, S., Vhaduri, S., & Lin, C.-H. (2026) · Artificial Intelligence Review · doi:10.1007/s10462-026-11670-y
VQ-CycleDiffusion: Vector Quantized Cycle-Consistent Diffusion Models for Voice Conversion
Yook, D., Kim, S., & Chang, H.-P. (2026) · Applied Sciences · doi:10.3390/app16189314
Multimodal Style Control in Voice Conversion Using Textual and Acoustic Prompts
Kim, M., Um, S., & Kang, H.-G. (2026) · IEEE Transactions on Audio Speech and Language Processing · doi:10.1109/taslpro.2026.3742217
“To address these limitations, non-parallel VC approaches have been proposed to remove the dependency on time-aligned speech pairs and enable training with unpaired data [5], [6], [7], [8], [9].”
CycleGAN-Based Representation-Level Translation of Asian-Accented English Speech Features
Pendyala, V. S., & Kothadia, K. (2026) · Electronics · doi:10.3390/electronics15204676
Defending Against Audio Spoofing Attacks via Emotion-Aware Prototype Metric Learning
Li, X., Gong, Y., Zou, D., & Chen, J. (2026) · Lecture notes in computer science · doi:10.1007/978-981-92-6834-4_25
SSMamba-VC: Learning Linear-Time Sequence Modeling with Selective State Spaces for Voice Conversion
Purohit, R. M., Wade, K. S., & Patil, H. A. (2026) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn69854.2026.11729097
“Earlier efforts in non-parallel VC used Generative Adversarial Networks (GANs) to mitigate the spectral oversmoothing caused by Mean Squared Error [1], [2].”
Metadata
- Authors
- Kaneko, T., & Kameoka, H.
- Affiliations
- NTT (Japan) (JP)
- Venue
- 2018 26th European Signal Processing Conference (EUSIPCO)
- Publisher
- IEEE
- Published
- 2018-09
- Pages
- 2100-2104
- OpenAlex
- W2902070858
- Cited by
- 358
- References
- 61
- Citations by year
- 2026: 19 · 2025: 36 · 2024: 50 · 2023: 66 · 2022: 58 · 2021: 63 · 2020: 47 · 2019: 18 · 2018: 1
- FWCI
- 13.16
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- European Signal Processing Conference (EUSIPCO) (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Speech Recognition and Synthesis
- Physical Sciences › Computer Science › Signal Processing › Speech and Audio Processing
- Health Sciences › Health Professions › Speech and Hearing › Voice and Speech Disorders
- Keywords
- voice conversion, non-parallel voice conversion, CycleGAN, gated convolutional neural network, pseudo data generation, over-smoothing, modulation spectra
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.