CycleGAN-VC: Non-parallel Voice Conversion Using Cycle-Consistent Adversarial Networks

By Takuhiro Kaneko, Hirokazu Kameoka. European Signal Processing Conference, 2018

325 citations recorded by the source corpus.

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).

Subfields: Artificial Intelligence, Signal Processing

View Paper: https://www.semanticscholar.org/paper/2a6a33645431151eb70cfb77c0367278d6f9c1da

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Xu Tan
2. Tao Qin
3. Tie-Yan Liu
4. Frank K. Soong
5. Sheng Zhao
6. Hung-yi Lee
7. Junichi Yamagishi
8. Yu Zhang

In Signal Processing

1. Xu Tan
2. Tao Qin
3. Frank K. Soong
4. Tie-Yan Liu
5. Sheng Zhao
6. Hung-yi Lee
7. Junichi Yamagishi
8. Haizhou Li

Citations in the corpus

• Emotional Voice Conversion: Theory, Databases and ESD (2021, background): “he cycle-consistency loss eliminates the need for parallel training ata through circular conversion, and the identity-mapping loss helps reserve the linguistic content between the input and output (Kaneko nd Kameoka, 2017; Fang et al., 2018; Kaneko and Kameoka, 2018).”
• An Overview of Affective Speech Synthesis and Conversion in the Deep Learning Era (2022, methodology): “However, despite several attempts, no notable progress was made until recent years, which saw the advent of DNNs. Recently proposed methods exploit the representation power of DNNs by means of VAEs [92], [93], GANs [94], [95], and seq2seq models [96], [97].”
• Unsupervised Learning of Disentangled Speech Content and Style Representation (2020, methodology): “For example [12, 13] used CycleGAN [14] to perform voice conversion without any paired data.”
• Axial Residual Networks for CycleGAN-based Voice Conversion (2021, methodology): “Many strategies have been proposed to this end: sequence-to-sequence architectures that explicitly decompose linguistic content via automatic speech recognition (ASR) and text-to-speech (TTS) [2, 3], variational autoencoder(VAE)based methods [4, 5], the style-changing StarGAN [6] applied to the doma”
• An Overview of Voice Conversion and Its Challenges: From Statistical Modeling to Deep Learning (2020, methodology): “Combining these three loss functions, we can obtain the overall loss function of CycleGAN [48], [210].”
• How Far Are We from Robust Voice Conversion: A Survey (2020, background): “Recently, several models were proposed to utilize non-parallel data [1, 2, 3, 4, 5].”
• MBNET: MOS Prediction for Synthesized Speech with Mean-Bias Network (2021, background): “Speech quality assessment [1] aims to measure the quality of synthesized speech and has long been a challenge in speech synthesis areas such as text to speech [2, 3, 4] and voice conversion [5, 6].”
• Deepfakes generation and detection: state-of-the-art, open challenges, countermeasures, and way forward (2021, methodology): “Recent GAN-based approaches, such as CycleGAN [253–256], VAW-GAN [257], and StarGAN [258], attempt to achieve high-quality transformed speech using non-parallel training data.”