A Multidomain Generative Adversarial Network for Hoarse-to-Normal Voice Conversion

This paper proposes a non-parallel voice conversion (VC) method using a variant of the conditional variational autoencoder (VAE) called an auxiliary classifier VAE. The proposed method has two key features. First, it adopts fully convolutional architectures to construct the encoder and decoder networks so that the networks can learn conversion rules that capture the time dependencies in the acoustic feature sequences of source and target speech. Second, it uses information-theoretic regularization for the model training to ensure that the information in the attribute class label will not be lost in the conversion process. With regular conditional VAEs, the encoder and decoder are free to ignore the attribute class label input. This can be problematic since in such a situation, the attribute class label will have little effect on controlling the voice characteristics of input speech at test time. Such situations can be avoided by introducing an auxiliary classifier and training the encoder and decoder so that the attribute classes of the decoder outputs are correctly predicted by the classifier. We also present several ways to convert the feature sequence of input speech using the trained encoder and decoder and compare them in terms of audio quality through objective and subjective evaluations. We confirmed experimentally that the proposed method outperformed baseline non-parallel VC systems and performed comparably to an open-source parallel VC system trained using a parallel corpus in a speaker identity conversion task.

Although voice conversion (VC) algorithms have achieved remarkable success along with the development of machine learning, superior performance is still difficult to achieve when using nonparallel data. In this paper, we propose using a cycle-consistent adversarial network (CycleGAN) for nonparallel data-based VC training. A CycleGAN is a generative adversarial network (GAN) originally developed for unpaired image-to-image translation. A subjective evaluation of inter-gender conversion demonstrated that the proposed method significantly outperformed a method based on the Merlin open source neural network speech synthesis system (a parallel VC system adapted for our setup) and a GAN-based parallel VC system. This is the first research to show that the performance of a nonparallel VC method can exceed that of state-of-the-art parallel VC methods.

This paper proposes a new voice conversion (VC) task from human speech to dog-like speech while preserving linguistic information as an example of human to non-human creature voice conversion (H2NH-VC) tasks. Although most VC studies deal with human to human VC, H2NH-VC aims to convert human speech into non-human creature-like speech. Non-parallel VC allows us to develop H2NH-VC, because we cannot collect a parallel dataset that non-human creatures speak human language. In this study, we propose to use dogs as an example of a non-human creature target domain and define the "speak like a dog" task. To clarify the possibilities and characteristics of the "speak like a dog" task, we conducted a comparative experiment using existing representative non-parallel VC methods in acoustic features (Mel-cepstral coefficients and Mel-spectrograms), network architectures (five different kernel-size settings), and training criteria (variational autoencoder (VAE)- based and generative adversarial network-based). Finally, the converted voices were evaluated using mean opinion scores: dog-likeness, sound quality and intelligibility, and character error rate (CER). The experiment showed that the employment of the Mel-spectrogram improved the dog-likeness of the converted speech, while it is challenging to preserve linguistic information. Challenges and limitations of the current VC methods for H2NH-VC are highlighted.

Voice conversion (VC) is a cutting-edge technology that enables the transformation of raw speech into high-quality audio resembling a target speaker, while preserving the original linguistic content and prosodic patterns.In this paper, we propose a novel voice conversion algorithm, Multi-Dimensional Perception Flow Matching (MPFM-VC).Unlike traditional approaches that directly generate waveform outputs, MPFM-VC models the evolutionary trajectory of mel-spectrograms through a flow matching framework, and incorporates a multi-dimensional feature perception network to enhance the stability and quality of speech synthesis. Additionally, we introduce a content perturbation method during training to improve the model’s generalization ability and reduce inference-time artifacts. To further increase speaker similarity, an adversarial training mechanism on speaker embeddings is employed to achieve effective disentanglement between content and speaker identity representations, thereby enhancing the timbre consistency of the converted speech. Experimental results on both speech and singing voice conversion tasks demonstrate that MPFM-VC outperforms existing state-of-the-art VC models in both subjective and objective evaluation metrics. The synthesized speech exhibits significantly improved naturalness, clarity, and timbre fidelity, validating the effectiveness of the proposed approach.

Singing voice conversion (SVC) is a task of converting the perception of the source speaker's identity to the target speaker without changing lyrics and rhythm. Recent approaches in traditional voice conversion involve the use of the generative models, such as Variational Autoencoders (VAE), and Generative Adversarial Networks (GANs). However, in the case of SVC, GANs are not explored much. The only system that has been proposed in the literature uses traditional GAN on the parallel data. The parallel data collection for real scenarios (with the same background music) is not feasible. Moreover, in the presence of background music, SVC is one of the most challenging tasks as it involves the source separation of vocals from the inputs, which will have some noise. Therefore, in this paper, we propose transfer learning, and fine-tuning-based Cycle consistent GAN (CycleGAN) model for non-parallel SVC, where music source separation is done using Deep Attractor Network (DANet). We designed seven different possible systems to identify the best possible combination of transfer learning and fine-tuning. Here, we use a more challenging database, MUSDB18, as our primary dataset, and we also use the NUS-48E database to pre-train CycleGAN. We perform extensive analysis via objective and subjective measures and report that with a 4.14 MOS score out of 5 for naturalness, the CycleGAN model pre-trained on NUS-48E corpus performs the best compared to the other systems described in the paper.

Voice conversion is a challenging task which transforms the voice characteristics of a source speaker to a target speaker without changing linguistic content. Recently, there have been many works on many-to-many Voice Conversion (VC) based on Variational Autoencoder (VAEs) achieving good results, however, these methods lack the ability to disentangle speaker identity and linguistic content to achieve good performance on unseen speaker scenarios. In this paper, we propose a new method based on feature disentanglement to tackle many to many voice conversion. The method has the capability to disentangle speaker identity and linguistic content from utterances, it can convert from many source speakers to many target speakers with a single autoencoder network. Moreover, it naturally deals with the unseen target speaker scenarios. We perform both objective and subjective evaluations to show the competitive performance of our proposed method compared with other state-of-the-art models in terms of naturalness and target speaker similarity.

In this paper, we present a voice conversion (VC) method that utilizes conditional restricted Boltzmann machines (CRBMs) for each speaker to obtain time-invariant speaker-independent spaces where voice features are converted more easily than those in an original acoustic feature space. First, we train two CRBMs for a source and target speaker independently using speaker-dependent training data (without the need to parallelize the training data). Then, a small number of parallel data are fed into each CRBM and the high-order features produced by the CRBMs are used to train a concatenating neural network (NN) between the two CRBMs. Finally, the entire network (the two CRBMs and the NN) is fine-tuned using the acoustic parallel data. Through voice-conversion experiments, we confirmed the high performance of our method in terms of objective and subjective evaluations, comparing it with conventional GMM, NN, and speaker-dependent DBN approaches.

Variational auto-encoder (VAE) is an effective neural network architecture to disentangle a speech utterance into speaker identity and linguistic content latent embeddings, then generate an utterance for a target speaker from that of a source speaker. This is possible by concatenating the identity embedding of the target speaker and the content embedding of the source speaker uttering a desired sentence. In this work, we propose to improve VAE models with self-attention and structural regularization (RGSM). Specifically, we found a suitable location of VAE’s decoder to add a self-attention layer for incorporating non-local information in generating a converted utterance and hiding the source speaker’s identity. We applied relaxed groupwise splitting method (RGSM) to regularize network weights and remarkably enhance generalization performance. In experiments of zero-shot many-to-many voice conversion task on VCTK data set, with the self-attention layer and relaxed group-wise splitting method, our model achieves a gain of speaker classification accuracy on unseen speakers by 28.3% while slightly improved conversion voice quality in terms of MOSNet scores. Our encouraging findings point to future research on integrating more variety of attention structures in VAE framework while controlling model size and overfitting for advancing zero-shot many-to-many voice conversions <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> . <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> The work was partially supported by NSF grants DMS-1854434 and DMS-1952644 at UC Irvine.

This paper proposes novel frameworks for non-parallel voice conversion (VC) using variational autoencoders (VAEs). Although conventional VAE-based VC models can be trained using non-parallel speech corpora with given speaker representations, phonetic contents of the converted speech tend to vanish because of an over-regularization issue often observed in latent variables of the VAEs. To overcome the issue, this paper proposes a VAE-based non-parallel VC conditioned by not only the speaker representations but also phonetic contents of speech represented as phonetic posteriorgrams (PPGs). Since the phonetic contents are given during the training, we can expect that the VC models effectively learn speaker-independent latent features of speech. Focusing on the point, this paper also extends the conventional VAE-based non-parallel VC to many-to-many VC that can convert arbitrary speakers' characteristics into another arbitrary speakers' ones. We investigate two methods to estimate speaker representations for speakers not included in speech corpora used for training VC models: 1) adapting conventional speaker codes, and 2) using d-vectors for the speaker representations. Experimental results demonstrate that 1) PPGs successfully improve both naturalness and speaker similarity of the converted speech, and 2) both speaker codes and d-vectors can be adopted to the VAE-based many-to-many non-parallel VC.

In conventional voice conversion methods, some features of a speech signal’s spectrum envelope are first extracted. Then, these features are converted so as to best match a target speaker’s speech by designing and using a set of conversions. Ultimately, the spectrum envelope of the target speaker’s speech signal is reconstructed from the converted features. The spectrum envelope reconstructed from the converted features usually deviates from its natural form. This aberration from the natural form observed in cases such as over-smoothing, over-fitting, and widening of formants is partially caused by two factors: (1) there is an error in the reconstruction of spectrum envelope from the features, and (2) the set of features extracted from the spectrum envelope of the speech signal is not closed. A method is put forward to improve the naturalness of speech by means of \(\epsilon \)-closed sets of extended vectors in voice conversion systems. In this approach, \(\epsilon \)-closed sets to reconstruct the natural spectrum envelope of a signal in the synthesis phase are introduced. The elements of these sets are generated by forming a group of extended vectors of features and applying a quantization scheme on the features of a speech signal. The use of this method in speech synthesis leads to a noticeable reduction of error in spectrum reconstruction from the features. Furthermore, the final spectrum envelope extracted from voice conversions maintains its natural form and, consequently, the problems arising from the deviation of voice from its natural state are resolved. The above method can be generally used as one phase of speech synthesis. It is independent of the voice conversion technique used and its parallel or non-parallel training method, and can be applied to improve the naturalness of the generated speech signal in all common voice conversion methods. Moreover, this method can be used in other fields of speech processing like texts to speech systems and vocoders to improve the quality of the output signal in the synthesis step.

Among the wide variety of image generative models, two models stand out: Variational Auto Encoders (VAE) and Generative Adversarial Networks (GAN). GANs can produce realistic images, but they suffer from mode collapse and do not provide simple ways to get the latent representation of an image. On the other hand, VAEs do not have these problems, but they often generate images less realistic than GANs. In this article, we explain that this lack of realism is partially due to a common underestimation of the natural image manifold dimensionality. To solve this issue we introduce a new framework that combines VAE and GAN in a novel and complementary way to produce an auto-encoding model that keeps VAEs properties while generating images of GAN-quality. We evaluate our approach both qualitatively and quantitatively on five image datasets.

Synthetic aperture radar (SAR) and optical sensing are two important means of Earth observation. SAR-to-optical image translation (S2OIT) can integrate the advantages of both and assist SAR image interpretation under all-day and all-weather conditions. The existing S2OIT methods generally follow a paired training paradigm, which is difficult when dealing with the unpaired S2OIT application scenarios. Moreover, the generator and discriminator in current S2OIT methods have insufficient scene memory for SAR images, resulting in regional landform deformation in the generated images. To address these issues, we propose a novel generative adversarial network capable of memorizing scene for unpaired S2OIT called MS-GAN. The cycle learning framework based on cycle generative adversarial network for unpaired S2OIT is designed to construct the translation mapping between unpaired SAR and optical images. The multiscale representation generator is constructed for multiscale fusion and utilization of scene features of SAR images. The proposed multireceptive field discriminator has the ability to enhance scene memory and generate higher quality optical images in different landforms. In addition, the designed subbands shrinkage denoising module can further suppress the effect of speckle noise in SAR images on the quality of the generated results. Extensive experiments conducted on three challenging datasets SEN1-2, WHU-SEN-City, and QXS-SAROPT demonstrate that the proposed MS-GAN outperforms the state-of-the-art methods on both subjective and objective evaluation metrics.

Voice conversion can be reduced to a problem to find a transformation function between the corresponding speech sequences of two speakers. Perhaps the most voice conversions methods are GMM-based statistical mapping methods [1, 2]. However, the classical GMM-based mapping is frame-to-frame, and cannot take account of the contextual information existing over a speech sequence. It is well known that HMM yields an efficient method to model the density of a whole speech sequence and has found great successes in speech recognition and synthesis. Inspired by this fact, this paper studies how to use HMM for voice conversion. We derive an HMM-based sequence-to-frame mapping function with statistical analysis. Different from previous HMM-based voice conversion methods [3, 4, 5] that used forced alignment for segmentation and transform frames aligned to a state with its associated linear transformation, our method has a soft mapping function as a weighted summation of linear transformations. The weights are calculated as the HMM posterior probabilities of frames. We also propose and compare two methods to learn the parameters of our mapping functions, namely least square error estimation and maximum likelihood estimation. We carried out experiments to examine the proposed HMM-based method for voice conversion.

Non-parallel data voice conversion (VC) have achieved considerable breakthroughs recently through introducing bottleneck features (BNFs) extracted by the automatic speech recognition(ASR) model. However, selection of BNFs have a significant impact on VC result. For example, when extracting BNFs from ASR trained with Cross Entropy loss (CE-BNFs) and feeding into neural network to train a VC system, the timbre similarity of converted speech is significantly degraded. If BNFs are extracted from ASR trained using Connectionist Temporal Classification loss (CTC-BNFs), the naturalness of the converted speech may decrease. This phenomenon is caused by the difference of information contained in BNFs. In this paper, we proposed an any-to-one VC method using hybrid bottleneck features extracted from CTC-BNFs and CE-BNFs to complement each other advantages. Gradient reversal layer and instance normalization were used to extract prosody information from CE-BNFs and content information from CTC-BNFs. Auto-regressive decoder and Hifi-GAN vocoder were used to generate high-quality waveform. Experimental results show that our proposed method achieves higher similarity, naturalness, quality than baseline method and reveals the differences between the information contained in CE-BNFs and CTC-BNFs as well as the influence they have on the converted speech.

Voice Conversion (VC) is a task of converting speaker-dependent features of a source speaker&#x0027;s speech without changing the linguistic content. There are many successful VC systems, each trying to overcome some challenges. These challenges include the unavailability of parallel data and solving problems due to the language difference between the source and target speech. Also, one of these challenges is extending the VC system to cover a conversion across many source and target domains with minimal cost. Generative Adversarial Networks (GANs) are showing promising VC results. This work focuses on exploring many-to-many non-parallel GAN-based mono-lingual VC models (nine models that are highly cited), explains the used evaluation methods including objective and subjective methods (eight evaluation methods are presented), and comments on these models.