Point clouds are widely used in various fields, including augmented reality (AR) and robotics, where relighting and texture editing are crucial for realistic visualization. Achieving these tasks requires accurately separating albedo from shade. However, performing this separation on point clouds presents two key challenges: (1) the non-grid structure of point clouds makes conventional image-based decomposition models ineffective, and (2) point-cloud models designed for other tasks do not explicitly consider global-light direction, resulting in inaccurate shade. In this paper, we introduce Intrinsic Point-Cloud Decomposition (IPCD), which extends image decomposition to the direct decomposition of colored point clouds into albedo and shade. To overcome challenge (1), we propose IPCD-Net that extends image-based model with point-wise feature aggregation for non-grid data processing. For challenge (2), we introduce Projection-based Luminance Distribution (PLD) with a hierarchical feature refinement, capturing global-light ques via multi-view projection. For comprehensive evaluation, we create a synthetic outdoor-scene dataset. Experimental results demonstrate that IPCD-Net reduces cast shadows in albedo and enhances color accuracy in shade. Furthermore, we showcase its applications in texture editing, relighting, and point-cloud registration under varying illumination. Finally, we verify the real-world applicability of IPCD-Net.

Estimating age from facial images is a fundamental task. In this task, age labels have ambiguity because faces of the same individual across similar ages are often difficult to distinguish. To model this ambiguity, label distribution learning (LDL) trains a deep neural network (DNN) using a label distribution, i.e., the probability that an image belongs to each age, instead of a single age label. However, the heuristic constraints utilized for LDL often fail to accurately model the label ambiguity. Therefore, we propose a novel LDL method called distribution highlighted reference-based LDL (DHRL), which introduces an input-dependent constraint by utilizing a reference DNN pre-trained with any LDL method and minimizing the gap between the reference and target DNNs’ outputs. DHRL incorporates two techniques to highlight the label ambiguity hidden in the pre-trained reference DNN’s output: noisy augmentation-based ensembling (NAE) and different scale multi-temperature (DSM). NAE inputs noisy images to the reference DNN and provides an ensemble effect by averaging all the outputs. DSM sets multiple temperatures simultaneously in the gap minimization between the two DNNs’ outputs. Experimental results indicate that our method achieves state-of-the-art performance across various datasets and conditions.

Recent advancements in neural 3D representations, such as neural radiance fields (NeRF) and 3D Gaussian splatting (3DGS), have made accurate estimation of the 3D structure from multiview images possible. However, this capability is limited to estimating the visible external structure, and it is still difficult to identify the invisible internal structure hidden behind the surface. To overcome this limitation, we address a new task called structure from collision (SfC), which aims to estimate the structure (including the invisible internal one) of an object from the appearance changes at collision. To solve this task, we propose a novel model called SfC-NeRF, which optimizes the invisible internal structure of the object through a video sequence under physical, appearance (i.e., visible external structure)-preserving, and keyframe constraints. In particular, to avoid falling into undesirable local optima owing to its ill-posed nature, we propose volume annealing, i.e., searching for the global optima by repeatedly reducing and expanding the volume. Extensive experiments on 115 objects involving diverse structures (i.e., various cavity shapes, locations, and sizes) and various material properties reveal the properties of SfC and demonstrate the effectiveness of the proposed SfC-NeRF.1

This study addresses the task of speech quality assessment (SQA), which aims to automatically predict the subjective quality of a given speech. Recent efforts have focused on training neural-based models to predict the mean opinion score (MOS) of speech samples produced by text-to-speech or voice conversion systems. We aim to enhance the performance of the models by a score aggregation method instead of MOS. The proposed method mitigates the effects of some issues arising from constraints imposed by limited options. Our method assumes annotators internally consider continuous scores and pick the nearest discrete rating. By modeling this process, we approximate the rating distribution by quantizing the latent continuous distribution. We then use the peak of the latent distribution, estimated through the loss between the distribution and actual ratings, as the new value instead of MOS. Experimental results demonstrate that substituting MOSNet’s target with this proposed value improves prediction performance.

Intrinsic image decomposition (IID) is the task of separating an image into albedo and shade. In real-world scenes, it is difficult to quantitatively assess IID quality due to the unavailability of ground truth. The existing method provides the relative reflection intensities based on human-judged annotations. However, these annotations have challenges in subjectivity, relative evaluation, and hue non-assessment. To address these, we propose a concept of quantitative evaluation with a calculated albedo from a hyperspectral imaging and light detection and ranging (LiDAR) intensity. Additionally, we introduce an optional albedo densification approach based on spectral similarity. This paper conducted a concept verification in a laboratory environment, and suggested the feasibility of an objective, absolute, and hue-aware assessment.1

Unsupervised intrinsic image decomposition (IID) is the task of separating a natural image into albedo and shade without ground truth during training. Although a recent model employing light detection and ranging (LiDAR) intensity demonstrated impressive performance, the necessity of LiDAR intensity during inference restricts its practicality. To expand the usage scenario while maintaining the IID quality achieved by using both an image and its corresponding LiDAR intensity, we propose a novel approach that utilizes an image without LiDAR intensity during inference while utilizing both an image and LiDAR intensity during training. Specifically, our proposed model processes an image and LiDAR intensity individually using distinct encoder paths during training, but utilizes only an imageencoder path during inference. Additionally, we introduce an albedo-alignment loss aligning the gray-scale albedo from an image to that from its corresponding LiDAR intensity. LiDAR intensity is not affected by illumination effects including cast shadows, thus albedo-alignment loss transfers the illumination-invariant property of LiDAR intensity to the image-encoder path. Furthermore, we also propose image-LiDAR conversion (ILC) paths that mutually translates the style of an image and LiDAR intensity. IID models translate an image into albedo and shade styles while keeping the image contents, thus it is important to separate the image into contents and style. Trained with pairs of an image and its corresponding LiDAR intensity which share contents but differ in style, the mutual translation in ILC paths improve the accuracy of the separation. Consequently, our model achieves comparable IID quality to the existing model with LiDAR intensity, while utilizing only an image without LiDAR intensity during inference.

Geometry-agnostic system identification is a technique for identifying the geometry and physical properties of an object from video sequences without any geometric as-sumptions. Recently, physics-augmented continuum neural radiance fields (PAC-NeRF) has demonstrated promising results for this technique by utilizing a hybrid Eulerian-Lagrangian representation, in which the geometry is represented by the Eulerian grid representations of NeRF, the physics is described by a material point method (MPM), and they are connected via Lagrangian particles. However, a notable limitation of PAC-NeRF is that its performance is sensitive to the learning of the geometry from the first frames owing to its two-step optimization. First, the grid representations are optimized with the first frames of video sequences, and then the physical properties are optimized through video sequences utilizing the fixed first-frame grid representations. This limitation can be critical when learning of the geometric structure is difficult, for example, in a few-shot (sparse view) setting. To overcome this limitation, we propose Lagrangian particle optimization (LPO), in which the positions and features of particles are opti-mized through video sequences in Lagrangian space. This method allows for the optimization of the geometric structure across the entire video sequence within the physical constraints imposed by the MPM. The experimental results demonstrate that the LPO is useful for geometric correction and physical identification in sparse-view settings.11Video samples are available at https://www.keel.ntt.co.jp/people/kaneko.takuhiro/projeets/lpo/

A generative adversarial network (GAN)-based vocoder trained with an adversarial discriminator is commonly used for speech synthesis because of its fast, lightweight, and high-quality characteristics. However, this data-driven model requires a large amount of training data incurring high data-collection costs. This fact motivates us to train a GAN-based vocoder on limited data. A promising solution is to augment the training data to avoid overfitting. However, a standard discriminator is unconditional and insensitive to distributional changes caused by data augmentation. Thus, augmented speech (which can be extraordinary) may be considered real speech. To address this issue, we propose an augmentation-conditional discriminator (AugCondD) that receives the augmentation state as input in addition to speech, thereby assessing input speech according to augmentation state, without inhibiting the learning of the original non-augmented distribution. Experimental results indicate that AugCondD improves speech quality under limited data conditions while achieving comparable speech quality under sufficient data conditions.1

The automatic speech quality assessment (SQA) has been extensively studied to predict the speech quality without time-consuming questionnaires. Recently, neural-based SQA models have been actively developed for speech samples produced by text-to-speech or voice conversion, with a primary focus on training mean opinion score (MOS) prediction models. The quality of each speech sample may not be consistent across the entire duration, and it remains unclear which segments of the speech receive the primary focus from humans when assigning subjective evaluation for MOS calculation. We hypothesize that when humans rate speech, they tend to assign more weight to low-quality speech segments, and the variance in ratings for each sample is mainly due to accidental assignment of higher scores when overlooking the poor quality speech segments. Motivated by the hypothesis, we analyze the VCC2018 and BVCC datasets. Based on the hypothesis, we propose the more reliable representative value Nlow-MOS, the mean of the N-lowest opinion scores. Our experiments show that LCC and SRCC improve compared to regular MOS when employing Nlow-MOS to MOSNet training. This result suggests that Nlow-MOS is a more intrinsic representative value of subjective speech quality and makes MOSNet a better comparator of VC models.

In this paper, we propose a non-parallel any-to-many voice conversion (VC) method termedVoiceGrad. Inspired by WaveGrad, a recently introduced novel waveform generation method, VoiceGrad is based upon the concepts of score matching, Langevin dynamics, and diffusion models. The idea involves training a score approximator, a fully convolutional network with a U-Net structure, to predict the gradient of the log density of the speech feature sequences of multiple speakers. The trained score approximator can be used to perform VC by using annealed Langevin dynamics or reverse diffusion process to iteratively update an input feature sequence towards the nearest stationary point of the target distribution. Thanks to the nature of this concept, VoiceGrad enables any-to-many VC, a VC scenario in which the speaker of input speech can be arbitrary, and allows for non-parallel training, which requires no parallel utterances.

Intrinsic image decomposition (IID) is the task that decomposes a natural image into albedo and shade. While IID is typically solved through supervised learning methods, it is not ideal due to the difficulty in observing ground truth albedo and shade in general scenes. Conversely, unsupervised learning methods are currently underperforming supervised learning methods since there are no criteria for solving the ill-posed problems. Recently, light detection and ranging (LiDAR) is widely used due to its ability to make highly precise distance measurements. Thus, we have focused on the utilization of LiDAR, especially LiDAR intensity, to address this issue. In this paper, we propose unsupervised intrinsic image decomposition with LiDAR intensity (IID-LI). Since the conventional unsupervised learning methods consist of image-to-image transformations, simply inputting LiDAR intensity is not an effective approach. Therefore, we design an intensity consistency loss that computes the error between LiDAR intensity and gray-scaled albedo to provide a criterion for the ill-posed problem. In addition, LiDAR intensity is difficult to handle due to its sparsity and occlusion, hence, a LiDAR intensity densification module is proposed. We verified the estimating quality using our own dataset, which include RGB images, LiDAR intensity and human judged annotations. As a result, we achieved an estimation accuracy that outperforms conventional unsupervised learning methods.

In speech synthesis, a generative adversarial network (GAN), training a generator (speech synthesizer) and a discriminator in a min-max game, is widely used to improve speech quality. An ensemble of discriminators is commonly used in recent neural vocoders (e.g., HiFi-GAN) and end-to-end text-to-speech (TTS) systems (e.g., VITS) to scrutinize waveforms from multiple perspectives. Such discriminators allow synthesized speech to adequately approach real speech; however, they require an increase in the model size and computation time according to the increase in the number of discriminators. Alternatively, this study proposes a Wave-U-Net discriminator, which is a single but expressive discriminator with Wave-U-Net architecture. This discriminator is unique; it can assess a waveform in a sample-wise manner with the same resolution as the input signal, while extracting multilevel features via an encoder and decoder with skip connections. This architecture provides a generator with sufficiently rich information for the synthesized speech to be closely matched to the real speech. During the experiments, the proposed ideas were applied to a representative neural vocoder (HiFi-GAN) and an end-to-end TTS system (VITS). The results demonstrate that the proposed models can achieve comparable speech quality with a 2.31 times faster and 14.5 times more lightweight discriminator when used in HiFi-GAN and a 1.90 times faster and 9.62 times more lightweight discriminator when used in VITS.1

This paper proposes a variational autoencoder (VAE)-based method for voice conversion (VC) on arbitrary source-target speaker pairs without parallel corpora, i.e., non-parallel any-to-any VC. One typical approach is to use speaker embeddings obtained from a speaker verification (SV) model as the condition for a VC model. However, converted speech is not guaranteed to reflect a target speaker’s characteristics in a naive combination of VC and SV models. Moreover, speaker embeddings are not designed for VC problems, leading to suboptimal conversion performance. To address these issues, the proposed method, JSV-VC, trains both VC and SV models jointly. The VC model is trained so that converted speech is verified as the target speaker in the SV model, while the SV model is trained in order to output consistent embeddings before and after the VC model. The experimental evaluation reveals that JSV-VC outperforms conventional any-to-any VC methods quantitatively and qualitatively.

Neural radiance fields (NeRFs) have shown impressive results for novel view synthesis. However, they depend on the repetitive use of a single-input single-output multilayer perceptron (SISO MLP) that maps 3D coordinates and view direction to the color and volume density in a sample-wise manner, which slows the rendering. We propose a multi-input multi-output NeRF (MIMO-NeRF) that reduces the number of MLPs running by replacing the SISO MLP with a MIMO MLP and conducting mappings in a group-wise manner. One notable challenge with this approach is that the color and volume density of each point can differ according to a choice of input coordinates in a group, which can lead to some notable ambiguity. We also propose a self-supervised learning method that regularizes the MIMO MLP with multiple fast reformulated MLPs to alleviate this ambiguity without using pretrained models. The results of a comprehensive experimental evaluation including comparative and ablation studies are presented to show that MIMO-NeRF obtains a good trade-off between speed and quality with a reasonable training time. We then demonstrate that MIMO-NeRF is compatible with and complementary to previous advancements in NeRFs by applying it to two representative fast NeRFs, i.e., a NeRF with a sampling network (DONeRF) and a NeRF with alternative representations (TensoRF).1

Fully unsupervised 3D representation learning has gained attention owing to its advantages in data collection. A successful approach involves a viewpoint-aware approach that learns an image distribution based on generative models (e.g., generative adversarial networks (GANs)) while generating various view images based on 3D-aware models (e.g., neural radiance fields (NeRFs)). However, they require images with various views for training, and consequently, their application to datasets with few or limited viewpoints remains a challenge. As a complementary approach, an aperture rendering GAN (AR-GAN) that employs a defocus cue was proposed. However, an AR-GAN is a CNN-based model and represents a defocus independently from a viewpoint change despite its high correlation, which is one of the reasons for its performance. As an alternative to an AR-GAN, we propose an aperture rendering NeRF (AR-NeRF), which can utilize viewpoint and defocus cues in a unified manner by representing both factors in a common ray-tracing framework. Moreover, to learn defocus-aware and defocus-independent representations in a disentangled manner, we propose aperture randomized training, for which we learn to generate images while randomizing the aperture size and latent codes independently. During our experiments, we applied AR-NeRF to various natural image datasets, including flower, bird, and face images, the results of which demonstrate the utility of AR-NeRF for un-supervised learning of the depth and defocus effects.

In recent text-to-speech synthesis and voice conversion systems, a mel-spectrogram is commonly applied as an intermediate representation, and the necessity for a mel-spectrogram vocoder is increasing. A mel-spectrogram vocoder must solve three inverse problems: recovery of the original-scale magnitude spectrogram, phase reconstruction, and frequency-to-time conversion. A typical convolutional mel-spectrogram vocoder solves these problems jointly and implicitly using a convolutional neural network, including temporal upsampling layers, when directly calculating a raw waveform. Such an approach allows skipping redundant processes during waveform synthesis (e.g., the direct reconstruction of high-dimensional original-scale spectrograms). By contrast, the approach solves all problems in a black box and cannot effectively employ the time-frequency structures existing in a mel-spectrogram. We thus propose iSTFTNet, which replaces some output-side layers of the mel-spectrogram vocoder with the inverse short-time Fourier transform (iSTFT) after sufficiently reducing the frequency dimension using upsampling layers, reducing the computational cost from black-box modeling and avoiding redundant estimations of high-dimensional spectrograms. During our experiments, we applied our ideas to three HiFi-GAN variants and made the models faster and more lightweight with a reasonable speech quality.1

This paper describes a method for distilling a recurrent-based sequence-to-sequence (S2S) voice conversion (VC) model. Although the performance of recent VCs is becoming higher quality, streaming conversion is still a challenge when considering practical applications. To achieve streaming VC, the conversion model needs a streamable structure, a causal layer rather than a non-causal layer. Motivated by this constraint and recent advances in S2S learning, we apply the teacher-student framework to recurrent-based S2S- VC models. A major challenge is how to minimize degradation due to the use of causal layers which masks future input information. Experimental evaluations show that except for male-to-female speaker conversion, our approach is able to maintain the teacher model's performance in terms of subjective evaluations despite the streamable student model structure. Audio samples can be accessed on http://www.kecl.ntt.co.jp/people/tanaka.ko/projects/dists2svc.

Understanding the 3D world from 2D projected natural images is a fundamental challenge in computer vision and graphics. Recently, an unsupervised learning approach has garnered considerable attention owing to its advantages in data collection. However, to mitigate training limitations, typical methods need to impose assumptions for viewpoint distribution (e.g., a dataset containing various viewpoint images) or object shape (e.g., symmetric objects). These assumptions often restrict applications; for instance, the application to non-rigid objects or images captured from similar viewpoints (e.g., flower or bird images) remains a challenge. To complement these approaches, we propose aperture rendering generative adversarial networks (AR-GANs), which equip aperture rendering on top of GANs, and adopt focus cues to learn the depth and depth-of-field (DoF) effect of unlabeled natural images. To address the ambiguities triggered by unsupervised setting (i.e., ambiguities between smooth texture and out-of-focus blurs, and between foreground and background blurs), we develop DoF mixture learning, which enables the generator to learn real image distribution while generating diverse DoF images. In addition, we devise a center focus prior to guiding the learning direction. In the experiments, we demonstrate the effectiveness of AR-GANs in various datasets, such as flower, bird, and face images, demonstrate their portability by incorporating them into other 3D representation learning GANs, and validate their applicability in shallow DoF rendering.

Generative adversarial networks (GANs) have gained considerable attention owing to their ability to reproduce images. However, they can recreate training images faithfully despite image degradation in the form of blur, noise, and compression, generating similarly degraded images. To solve this problem, the recently proposed noise robust GAN (NR-GAN) provides a partial solution by demonstrating the ability to learn a clean image generator directly from noisy images using a two-generator model comprising image and noise generators. However, its application is limited to noise, which is relatively easy to decompose owing to its additive and reversible characteristics, and its application to irreversible image degradation, in the form of blur, compression, and combination of all, remains a challenge. To address these problems, we propose blur, noise, and compression robust GAN (BNCR-GAN) that can learn a clean image generator directly from degraded images without knowledge of degradation parameters (e.g., blur kernel types, noise amounts, or quality factor values). Inspired by NR-GAN, BNCR-GAN uses a multiple-generator model composed of image, blur-kernel, noise, and quality-factor generators. However, in contrast to NR-GAN, to address irreversible characteristics, we introduce masking architectures adjusting degradation strength values in a data-driven manner using bypasses before and after degradation. Furthermore, to suppress uncertainty caused by the combination of blur, noise, and compression, we introduce adaptive consistency losses imposing consistency between irreversible degradation processes according to the degradation strengths. We demonstrate the effectiveness of BNCR-GAN through large-scale comparative studies on CIFAR-10 and a generality analysis on FFHQ. In addition, we demonstrate the applicability of BNCR-GAN in image restoration.

Non-parallel voice conversion (VC) is a technique for training voice converters without a parallel corpus. Cycle-consistent adversarial network-based VCs (CycleGAN-VC and CycleGAN-VC2) are widely accepted as benchmark methods. However, owing to their insufficient ability to grasp time-frequency structures, their application is limited to mel-cepstrum conversion and not mel-spectrogram conversion despite recent advances in mel-spectrogram vocoders. To overcome this, CycleGAN-VC3, an improved variant of CycleGAN-VC2 that incorporates an additional module called time-frequency adaptive normalization (TFAN), has been proposed. However, an increase in the number of learned parameters is imposed. As an alternative, we propose MaskCycleGAN-VC, which is another extension of CycleGAN-VC2 and is trained using a novel auxiliary task called filling in frames (FIF). With FIF, we apply a temporal mask to the input mel-spectrogram and encourage the converter to fill in missing frames based on surrounding frames. This task allows the converter to learn time-frequency structures in a self-supervised manner and eliminates the need for an additional module such as TFAN. A subjective evaluation of the naturalness and speaker similarity showed that MaskCycleGAN-VC outperformed both CycleGAN-VC2 and CycleGAN-VC3 with a model size similar to that of CycleGAN-VC2.1

This paper proposes a voice conversion (VC) method based on a sequence-to-sequence (S2S) learning framework, which enables simultaneous conversion of the voice characteristics, pitch contour, and duration of input speech. We previously proposed an S2S-based VC method using a transformer network architecture called the voice transformer network (VTN). The original VTN was designed to learn only a mapping of speech feature sequences from one speaker to another. Here, the main idea we propose is an extension of the original VTN that can simultaneously learn mappings among multiple speakers. This extension, called the many-to-many VTN, enables us to fully use available training data collected from multiple speakers by capturing common latent features that can be shared across different speakers. It also allows us to introduce a training loss called the identity mapping loss to ensure that the input feature sequence will remain unchanged when the source and target speaker indices are the same. Using this particular loss for model training has been found to be extremely effective in improving the performance of the model at test time. We conducted speaker identity conversion experiments and found that our model obtained higher sound quality and speaker similarity than baseline methods. We also found that our model, with a slight modification to its architecture, can handle any-to-many conversion tasks reasonably well.

Generative adversarial networks (GANs) are neural networks that learn data distributions through adversarial training. In intensive studies, recent GANs have shown promising results for reproducing training images. However, in spite of noise, they reproduce images with fidelity. As an alternative, we propose a novel family of GANs called noise robust GANs (NR-GANs), which can learn a clean image generator even when training images are noisy. In particular, NR-GANs can solve this problem without having complete noise information (e.g., the noise distribution type, noise amount, or signal-noise relationship). To achieve this, we introduce a noise generator and train it along with a clean image generator. However, without any constraints, there is no incentive to generate an image and noise separately. Therefore, we propose distribution and transformation constraints that encourage the noise generator to capture only the noise-specific components. In particular, considering such constraints under different assumptions, we devise two variants of NR-GANs for signal-independent noise and three variants of NR-GANs for signal-dependent noise. On three benchmark datasets, we demonstrate the effectiveness of NR-GANs in noise robust image generation. Furthermore, we show the applicability of NR-GANs in image denoising. Our code is available at https://github.com/takuhirok/NR-GAN/.

Non-parallel voice conversion (VC) is a technique for learning mappings between source and target speeches without using a parallel corpus.Recently, cycle-consistent adversarial network (CycleGAN)-VC and CycleGAN-VC2 have shown promising results regarding this problem and have been widely used as benchmark methods.However, owing to the ambiguity of the effectiveness of CycleGAN-VC/VC2 for mel-spectrogram conversion, they are typically used for mel-cepstrum conversion even when comparative methods employ mel-spectrogram as a conversion target.To address this, we examined the applicability of CycleGAN-VC/VC2 to mel-spectrogram conversion.Through initial experiments, we discovered that their direct applications compromised the time-frequency structure that should be preserved during conversion.To remedy this, we propose CycleGAN-VC3, an improvement of CycleGAN-VC2 that incorporates time-frequency adaptive normalization (TFAN).Using TFAN, we can adjust the scale and bias of the converted features while reflecting the time-frequency structure of the source mel-spectrogram.We evaluated CycleGAN-VC3 on inter-gender and intra-gender non-parallel VC.A subjective evaluation of naturalness and similarity showed that for every VC pair, CycleGAN-VC3 outperforms or is competitive with the two types of CycleGAN-VC2, one of which was applied to mel-cepstrum and the other to mel-spectrogram. 1

We previously proposed a method that allows for nonparallel voice conversion (VC) by using a variant of generative adversarial networks (GANs) called StarGAN. The main features of our method, called StarGAN-VC, are as follows: First, it requires no parallel utterances, transcriptions, or time alignment procedures for speech generator training. Second, it can simultaneously learn mappings across multiple domains using a single generator network and thus fully exploit available training data collected from multiple domains to capture latent features that are common to all the domains. Third, it can generate converted speech signals quickly enough to allow real-time implementations and requires only several minutes of training examples to generate reasonably realistic-sounding speech. In this article, we describe three formulations of StarGAN, including a newly introduced novel StarGAN variant called “Augmented classifier StarGAN (A-StarGAN)”, and compare them in a nonparallel VC task. We also compare them with several baseline methods.

This article proposes a voice conversion (VC) method using sequence-to-sequence (seq2seq or S2S) learning, which flexibly converts not only the voice characteristics but also the pitch contour and duration of input speech. The proposed method, called ConvS2S-VC, has three key features. First, it uses a model with a fully convolutional architecture. This is particularly advantageous in that it is suitable for parallel computations using GPUs. It is also beneficial since it enables effective normalization techniques such as batch normalization to be used for all the hidden layers in the networks. Second, it achieves many-to-many conversion by simultaneously learning mappings among multiple speakers using only a single model instead of separately learning mappings between each speaker pair using a different model. This enables the model to fully utilize available training data collected from multiple speakers by capturing common latent features that can be shared across different speakers. Owing to this structure, our model works reasonably well even without source speaker information, thus making it able to handle any-to-many conversion tasks. Third, we introduce a mechanism, called the conditional batch normalization that switches batch normalization layers in accordance with the target speaker. This particular mechanism has been found to be extremely effective for our many-to-many conversion model. We conducted speaker identity conversion experiments and found that ConvS2S-VC obtained higher sound quality and speaker similarity than baseline methods. We also found from audio examples that it could perform well in various tasks including emotional expression conversion, electrolaryngeal speech enhancement, and English accent conversion.

Generative adversarial networks (GANs) are a framework that learns a generative distribution through adversarial training. Recently, their class conditional extensions (e.g., conditional GAN (cGAN) and auxiliary classifier GAN (AC-GAN)) have attracted much attention owing to their ability to learn the disentangled representations and to improve the training stability. However, their training requires the availability of large-scale accurate class-labeled data, which are often laborious or impractical to collect in a real-world scenario. To remedy this, we propose a novel family of GANs called label-noise robust GANs (rGANs), which, by incorporating a noise transition model, can learn a clean label conditional generative distribution even when training labels are noisy. In particular, we propose two variants: rAC-GAN, which is a bridging model between AC-GAN and the label-noise robust classification model, and rcGAN, which is an extension of cGAN and solves this problem with no reliance on any classifier. In addition to providing the theoretical background, we demonstrate the effectiveness of our models through extensive experiments using diverse GAN configurations, various noise settings, and multiple evaluation metrics (in which we tested 402 conditions in total).

Non-parallel voice conversion (VC) is a technique for learning the mapping from source to target speech without relying on parallel data. This is an important task, but it has been challenging due to the disadvantages of the training conditions. Recently, CycleGAN-VC has provided a breakthrough and performed comparably to a parallel VC method without relying on any extra data, modules, or time alignment procedures. However, there is still a large gap between the real target and converted speech, and bridging this gap remains a challenge. To reduce the gap, we propose CycleGAN-VC2, which is an improved version of CycleGAN-VC incorporating three new techniques: an improved objective (two-step adversarial losses), improved generator (2-1-2D CNN), and improved discriminator (PatchGAN). We evaluated our method on a non-parallel VC task and analyzed the effect of each technique in detail. An objective evaluation showed that these techniques help bring the converted feature sequence closer to the target in terms of both global and local structures, which we assess by using Mel-cepstral distortion and modulation spectra distance, respectively. A subjective evaluation showed that CycleGAN-VC2 outperforms CycleGAN-VC in terms of naturalness and similarity for every speaker pair, including intra-gender and inter-gender pairs.

This paper describes a method based on a sequence-to-sequence learning (Seq2Seq) with attention and context preservation mechanism for voice conversion (VC) tasks. Seq2Seq has been outstanding at numerous tasks involving sequence modeling such as speech synthesis and recognition, machine translation, and image captioning. In contrast to current VC techniques, our method 1) stabilizes and accelerates the training procedure by considering guided attention and proposed context preservation losses, 2) allows not only spectral envelopes but also fundamental frequency contours and durations of speech to be converted, 3) requires no context information such as phoneme labels, and 4) requires no time-aligned source and target speech data in advance. In our experiment, the proposed VC framework can be trained in only one day, using only one GPU of an NVIDIA Tesla K80, while the quality of the synthesized speech is higher than that of speech converted by Gaussian mixture model-based VC and is comparable to that of speech generated by recurrent neural network-based text-to-speech synthesis, which can be regarded as an upper limit on VC performance.

Non-parallel multi-domain voice conversion (VC) is a technique for learning mappings among multiple domains without relying on parallel data.This is important but challenging owing to the requirement of learning multiple mappings and the nonavailability of explicit supervision.Recently, StarGAN-VC has garnered attention owing to its ability to solve this problem only using a single generator.However, there is still a gap between real and converted speech.To bridge this gap, we rethink conditional methods of StarGAN-VC, which are key components for achieving non-parallel multi-domain VC in a single model, and propose an improved variant called StarGAN-VC2.Particularly, we rethink conditional methods in two aspects: training objectives and network architectures.For the former, we propose a source-and-target conditional adversarial loss that allows all source domain data to be convertible to the target domain data.For the latter, we introduce a modulation-based conditional method that can transform the modulation of the acoustic feature in a domain-specific manner.We evaluated our methods on non-parallel multi-speaker VC.An objective evaluation demonstrates that our proposed methods improve speech quality in terms of both global and local structure measures.Furthermore, a subjective evaluation shows that StarGAN-VC2 outperforms StarGAN-VC in terms of naturalness and speaker similarity.

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.

This paper proposes the decision tree latent controller generative adversarial network (DTLC-GAN), an extension of a GAN that can learn hierarchically interpretable representations without relying on detailed supervision. To impose a hierarchical inclusion structure on latent variables, we incorporate a new architecture called the DTLC into the generator input. The DTLC has a multiple-layer tree structure in which the ON or OFF of the child node codes is controlled by the parent node codes. By using this architecture hierarchically, we can obtain the latent space in which the lower layer codes are selectively used depending on the higher layer ones. To make the latent codes capture salient semantic features of images in a hierarchically disentangled manner in the DTLC, we also propose a hierarchical conditional mutual information regularization and optimize it with a newly defined curriculum learning method that we propose as well. This makes it possible to discover hierarchically interpretable representations in a layer-by-layer manner on the basis of information gain by only using a single DTLC-GAN model. We evaluated the DTLC-GAN on various datasets, i.e., MNIST, CIFAR-10, Tiny ImageNet, 3D Faces, and CelebA, and confirmed that the DTLC-GAN can learn hierarchically interpretable representations with either unsupervised or weakly supervised settings. Furthermore, we applied the DTLC-GAN to image-retrieval tasks and showed its effectiveness in representation learning.

This paper proposes a method that allows non-parallel many-to-many voice conversion (VC) by using a variant of a generative adversarial network (GAN) called StarGAN. Our method, which we call StarGAN-VC, is noteworthy in that it (1) requires no parallel utterances, transcriptions, or time alignment procedures for speech generator training, (2) simultaneously learns many-to-many mappings across different attribute domains using a single generator network, (3) is able to generate converted speech signals quickly enough to allow real-time implementations and (4) requires only several minutes of training examples to generate reasonably realistic sounding speech. Subjective evaluation experiments on a non-parallel many-to-many speaker identity conversion task revealed that the proposed method obtained higher sound quality and speaker similarity than a state-of-the-art method based on variational autoencoding GANs.

We propose a learning-based filter that allows us to directly modify a synthetic speech waveform into a natural speech waveform. Speech-processing systems using a vocoder framework such as statistical parametric speech synthesis and voice conversion are convenient especially for a limited number of data because it is possible to represent and process interpretable acoustic features over a compact space, such as the fundamental frequency (F0) and mel-cepstrum. However, a well-known problem that leads to the quality degradation of generated speech is an over-smoothing effect that eliminates some detailed structure of generated/converted acoustic features. To address this issue, we propose a synthetic-to-natural speech waveform conversion technique that uses cycle-consistent adversarial networks and which does not require any explicit assumption about speech waveform in adversarial learning. In contrast to current techniques, since our modification is performed at the waveform level, we expect that the proposed method will also make it possible to generate “vocoder-less” sounding speech even if the input speech is synthesized using a vocoder framework. The experimental results demonstrate that our proposed method can 1) alleviate the over-smoothing effect of the acoustic features despite the direct modification method used for the waveform and 2) greatly improve the naturalness of the generated speech sounds.

We present a generative attribute controller (GAC), a novel functionality for generating or editing an image while intuitively controlling large variations of an attribute. This controller is based on a novel generative model called the conditional filtered generative adversarial network (CFGAN), which is an extension of the conventional conditional GAN (CGAN) that incorporates a filtering architecture into the generator input. Unlike the conventional CGAN, which represents an attribute directly using an observable variable (e.g., the binary indicator of attribute presence) so its controllability is restricted to attribute labeling (e.g., restricted to an ON or OFF control), the CFGAN has a filtering architecture that associates an attribute with a multi-dimensional latent variable, enabling latent variations of the attribute to be represented. We also define the filtering architecture and training scheme considering controllability, enabling the variations of the attribute to be intuitively controlled using typical controllers (radio buttons and slide bars). We evaluated our CFGAN on MNIST, CUB, and CelebA datasets and show that it enables large variations of an attribute to be not only represented but also intuitively controlled while retaining identity. We also show that the learned latent space has enough expressive power to conduct attribute transfer and attribute-based image retrieval.

We propose a postfilter based on a generative adversarial network (GAN) to compensate for the differences between natural speech and speech synthesized by statistical parametric speech synthesis. In particular, we focus on the differences caused by over-smoothing, which makes the sounds muffled. Over-smoothing occurs in the time and frequency directions and is highly correlated in both directions, and conventional methods based on heuristics are too limited to cover all the factors (e.g., global variance was designed only to recover the dynamic range). To solve this problem, we focus on “spectral texture”, i.e., the details of the time-frequency representation, and propose a learning-based postfilter that captures the structures directly from the data. To estimate the true distribution, we utilize a GAN composed of a generator and a discriminator. This optimizes the generator to produce samples imitating the dataset according to the adversarial discriminator. This adversarial process encourages the generator to fit the true data distribution, i.e., to generate realistic spectral texture. Objective evaluation of experimental results shows that the GAN-based postfilter can compensate for detailed spectral structures including modulation spectrum, and subjective evaluation shows that its generated speech is comparable to natural speech.

We propose a learning-based postfilter to reconstruct the high-fidelity spectral texture in short-term Fourier transform (STFT) spectrograms. In speech-processing systems, such as speech synthesis, voice conversion, and speech enhancement, the STFT spectrograms have been widely used as key acoustic representations. In these tasks, we normally need to precisely generate or predict the representations from inputs; however, generated spectra typically lack the fine structures close to the true data. To overcome these limitations and reconstruct spectra having finer structures, we propose a generative adversarial network (GAN)-based postfilter that is implicitly optimized to match the true feature distribution in adversarial learning. The challenge with this postfilter is that a GAN cannot be easily trained for very high-dimensional data such as the STFT. Therefore, we introduce a divide-and-concatenate strategy. We first divide the spectrograms into multiple frequency bands with overlap, train the GAN-based postfilter for the individual bands, and finally connect the bands with overlap. We applied our proposed postfilter to a DNN-based speech-synthesis task. The results show that our proposed postfilter can be used to reduce the gap between synthesized and target spectra, even in the highdimensional STFT domain.

While many studies in computer vision and pattern recognition have been actively conducted to recognize people's current states, few studies have tackled the problem of generating feedback on how people can improve their states, although there are many real-world applications such as in sports, education, and health care. In particular, it has been challenging to develop such a system that can adaptively generate feedback for real-world situations, namely various input and target states, since it requires formulating various rules of feedback to do so. We propose a learning-based method to solve this problem. If we can obtain a large amount of feedback annotations, it is possible to explicitly learn the rules, but it is difficult to do so due to the subjective nature of the task. To mitigate this problem, our method implicitly learns the rules from training data consisting of input images, key-point annotations, and state annotations that do not require professional knowledge in feedback. Given such training data, we first learn a multi-task deep neural network with state recognition and key-point localization. Then, we apply a novel propagation method for extracting feedback information from the network. We evaluated our method in a facial expression improvement task using real-world data and clarified its characteristics and effectiveness.