Music editing is an important step in music production, which has broad applications, including game development and film production. Most existing zero-shot text-guided editing methods rely on pretrained diffusion models by involving forward-backward diffusion processes. However, these methods often struggle to preserve the musical content. Additionally, text instructions alone usually fail to accurately describe the desired music. In this paper, we propose two music editing methods that improve the consistency between the original and edited music by leveraging score distillation. The first method, SteerMusic, is a coarse-grained zero-shot editing approach using delta denoising score. The second method, SteerMusic+, enables fine-grained personalized music editing by manipulating a concept token that represents a user-defined musical style. SteerMusic+ allows for the editing of music into user-defined musical styles that cannot be achieved by the text instructions alone. Experimental results show that our methods outperform existing approaches in preserving both music content consistency and editing fidelity. User studies further validate that our methods achieve superior music editing quality.

Music timbre transfer is a challenging task that involves modifying the timbral characteristics of an audio signal while preserving its melodic structure. In this paper, we propose a novel method based on dual diffusion bridges, trained using the CocoChorales Dataset, which consists of unpaired monophonic single-instrument audio data. Each diffusion model is trained on a specific instrument with a Gaussian prior. During inference, a model is designated as the source model to map the input audio to its corresponding Gaussian prior, and another model is designated as the target model to reconstruct the target audio from this Gaussian prior, thereby facilitating timbre transfer. We compare our approach against existing unsupervised timbre transfer models such as VAEGAN and Gaussian Flow Bridges (GFB). Experimental results demonstrate that our method achieves both better Fréchet Audio Distance (FAD) and melody preservation, as reflected by lower pitch distances (DPD) compared to VAEGAN and GFB. Additionally, we discover that the noise level from the Gaussian prior, σ, can be adjusted to control the degree of melody preservation and amount of timbre transferred.

Controllable generation through Stable Diffusion (SD) fine-tuning aims to improve fidelity, safety, and alignment with human guidance. Existing reinforcement learning from human feedback methods usually rely on predefined heuristic reward functions or pretrained reward models built on large-scale datasets, limiting their applicability to scenarios where collecting such data is costly or difficult. To effectively and efficiently utilize human feedback, we develop a framework, HERO, which leverages online human feedback collected on the fly during model learning. Specifically, HERO features two key mechanisms: (1) Feedback-Aligned Representation Learning, an online training method that captures human feedback and provides informative learning signals for fine-tuning, and (2) Feedback-Guided Image Generation, which involves generating images from SD's refined initialization samples, enabling faster convergence towards the evaluator's intent. We demonstrate that HERO is 4x more efficient in online feedback for body part anomaly correction compared to the best existing method. Additionally, experiments show that HERO can effectively handle tasks like reasoning, counting, personalization, and reducing NSFW content with only 0.5K online feedback. The code and project page are available at [https://hero-dm.github.io/](https://hero-dm.github.io/).

Diffusion models have seen notable success in continuous domains, leading to the development of discrete diffusion models (DDMs) for discrete variables. Despite recent advances, DDMs face the challenge of slow sampling speeds. While parallel sampling methods like $\tau$-leaping accelerate this process, they introduce _Compounding Decoding Error_ (CDE), where discrepancies arise between the true distribution and the approximation from parallel token generation, leading to degraded sample quality. In this work, we present _Jump Your Steps_ (JYS), a novel approach that optimizes the allocation of discrete sampling timesteps by minimizing CDE without extra computational cost. More precisely, we derive a practical upper bound on CDE and propose an efficient algorithm for searching for the optimal sampling schedule. Extensive experiments across image, music, and text generation show that JYS significantly improves sampling quality, establishing it as a versatile framework for enhancing DDM performance for fast sampling.

Sound content creation, essential for multimedia works such as video games and films, often involves extensive trial-and-error, enabling creators to semantically reflect their artistic ideas and inspirations, which evolve throughout the creation process, into the sound. Recent high-quality diffusion-based Text-to-Sound (T2S) generative models provide valuable tools for creators. However, these models often suffer from slow inference speeds, imposing an undesirable burden that hinders the trial-and-error process. While existing T2S distillation models address this limitation through $1$-step generation, the sample quality of $1$-step generation remains insufficient for production use. Additionally, while multi-step sampling in those distillation models improves sample quality itself, the semantic content changes due to their lack of deterministic sampling capabilities. Thus, developing a T2S generative model that allows creators to efficiently conduct trial-and-error while producing high-quality sound remains a key challenge. To address these issues, we introduce Sound Consistency Trajectory Models (SoundCTM), which allow flexible transitions between high-quality $1$-step sound generation and superior sound quality through multi-step deterministic sampling. This allows creators to efficiently conduct trial-and-error with $1$-step generation to semantically align samples with their intention, and subsequently refine sample quality with preserving semantic content through deterministic multi-step sampling. To develop SoundCTM, we reframe the CTM training framework, originally proposed in computer vision, and introduce a novel feature distance using the teacher network for a distillation loss. Additionally, while distilling classifier-free guided trajectories, we introduce a $\nu$-sampling, a new algorithm that offers another source of quality improvement. For the $\nu$-sampling, we simultaneously train both conditional and unconditional student models. For production-level generation, we scale up our model to 1B trainable parameters, making SoundCTM-DiT-1B the first large-scale distillation model in the sound community to achieve both promising high-quality $1$-step and multi-step full-band (44.1kHz) generation. Audio samples are available at \url{https://anonymus-soundctm.github.io/soundctm_iclr/}.

In typical multimodal contrastive learning, such as CLIP, encoders produce one point in the latent representation space for each input. However, one-point representation has difficulty in capturing the relationship and the similarity structure of a huge amount of instances in the real world. For richer classes of the similarity, we propose the use of weighted point sets, namely, sets of pairs of weight and vector, as representations of instances. In this work, we theoretically show the benefit of our proposed method through a new understanding of the contrastive loss of CLIP, which we call symmetric InfoNCE. We clarify that the optimal similarity that minimizes symmetric InfoNCE is the pointwise mutual information, and show an upper bound of excess risk on downstream classification tasks of representations that achieve the optimal similarity. In addition, we show that our proposed similarity based on weighted point sets consistently achieves the optimal similarity. To verify the effectiveness of our proposed method, we demonstrate pretraining of text-image representation models and classification tasks on common benchmarks.

Consistency Training (CT) has recently emerged as a strong alternative to diffusion models for image generation. However, non-distillation CT often suffers from high variance and instability, motivating ongoing research into its training dynamics. We propose Variational Consistency Training (VCT), a flexible and effective framework compatible with various forward kernels, including those in flow matching. Its key innovation is a learned noise-data coupling scheme inspired by Variational Autoencoders, where a data-dependent encoder models noise emission. This enables VCT to adaptively learn noise-to-data pairings, reducing training variance relative to the fixed, unsorted pairings in classical CT. Experiments on multiple image datasets demonstrate significant improvements: our method surpasses baselines, achieves state-of-the-art FID among non-distillation CT approaches on CIFAR-10, and matches SoTA performance on ImageNet 64x64 with only two sampling steps. Code is available at https://github.com/sony/vct.

By embedding discrete representations into a continuous latent space, we can leverage continuous-space latent diffusion models to handle generative modeling of discrete data. However, despite their initial success, most latent diffusion methods rely on fixed pretrained embeddings, limiting the benefits of joint training with the diffusion model. While jointly learning the embedding (via reconstruction loss) and the latent diffusion model (via score matching loss) could enhance performance, end-to-end training risks embedding collapse, degrading generation quality. To mitigate this issue, we introduce VQ-LCMD, a continuous-space latent diffusion framework within the embedding space that stabilizes training. VQ-LCMD uses a novel training objective combining the joint embedding-diffusion variational lower bound with a consistency-matching (CM) loss, alongside a shifted cosine noise schedule and random dropping strategy. Experiments on several benchmarks show that the proposed VQ-LCMD yields superior results on FFHQ, LSUN Churches, and LSUN Bedrooms compared to discrete-state latent diffusion models. In particular, VQ-LCMD achieves an FID of 6.81 for class-conditional image generation on ImageNet with 50 steps.

Restoring degraded music signals is essential to enhance audio quality for downstream music manipulation. Recent diffusion-based music restoration methods have demonstrated impressive performance, and among them, diffusion posterior sampling (DPS) stands out given its intrinsic properties, making it versatile across various restoration tasks. In this paper, we identify that there are potential issues which will degrade current DPS-based methods’ performance and introduce the way to mitigate the issues inspired by diverse diffusion guidance techniques including the RePaint (RP) strategy and the Pseudoinverse-Guided Diffusion Models (ΠGDM). We demonstrate our methods for the vocal declipping and bandwidth extension tasks under various levels of distortion and cutoff frequency, respectively. In both tasks, our methods outperform the current DPS-based music restoration benchmarks. We refer to http://carlosholivan.github.io/demos/audio-restoration-2023.html for examples of the restored audio samples.

Despite the recent advancements, conditional image generation still faces challenges of cost, generalizability, and the need for task-specific training. In this paper, we propose Manifold Preserving Guided Diffusion (MPGD), a training-free conditional generation framework that leverages pretrained diffusion models and off-the-shelf neural networks with minimal additional inference cost for a broad range of tasks. Specifically, we leverage the manifold hypothesis to refine the guided diffusion steps and introduce a shortcut algorithm in the process. We then propose two methods for on-manifold training-free guidance using pre-trained autoencoders and demonstrate that our shortcut inherently preserves the manifolds when applied to latent diffusion models. Our experiments show that MPGD is efficient and effective for solving a variety of conditional generation applications in low-compute settings, and can consistently offer up to 3.8× speed-ups with the same number of diffusion steps while maintaining high sample quality compared to the baselines.

Consistency Models (CM) (Song et al., 2023) accelerate score-based diffusion model sampling at the cost of sample quality but lack a natural way to trade-off quality for speed. To address this limitation, we propose Consistency Trajectory Model (CTM), a generalization encompassing CM and score-based models as special cases. CTM trains a single neural network that can -- in a single forward pass -- output scores (i.e., gradients of log-density) and enables unrestricted traversal between any initial and final time along the Probability Flow Ordinary Differential Equation (ODE) in a diffusion process. CTM enables the efficient combination of adversarial training and denoising score matching loss to enhance performance and achieves new state-of-the-art FIDs for single-step diffusion model sampling on CIFAR-10 (FID 1.73) and ImageNet at 64X64 resolution (FID 1.92). CTM also enables a new family of sampling schemes, both deterministic and stochastic, involving long jumps along the ODE solution trajectories. It consistently improves sample quality as computational budgets increase, avoiding the degradation seen in CM. Furthermore, unlike CM, CTM's access to the score function can streamline the adoption of established controllable/conditional generation methods from the diffusion community. This access also enables the computation of likelihood. The code is available at https://github.com/sony/ctm.

Generative adversarial networks (GANs) learn a target probability distribution by optimizing a generator and a discriminator with minimax objectives. This paper addresses the question of whether such optimization actually provides the generator with gradients that make its distribution close to the target distribution. We derive *metrizable conditions*, sufficient conditions for the discriminator to serve as the distance between the distributions, by connecting the GAN formulation with the concept of sliced optimal transport. Furthermore, by leveraging these theoretical results, we propose a novel GAN training scheme called the Slicing Adversarial Network (SAN). With only simple modifications, a broad class of existing GANs can be converted to SANs. Experiments on synthetic and image datasets support our theoretical results and the effectiveness of SAN as compared to the usual GANs. We also apply SAN to StyleGAN-XL, which leads to a state-of-the-art FID score amongst GANs for class conditional generation on CIFAR10 and ImageNet 256$\times$256. The code is available at https://github.com/sony/san.

The diffusion model performs remarkable in generating high-dimensional content but is computationally intensive, especially during training. We propose Progressive Growing of Diffusion Autoencoder (PaGoDA), a novel pipeline that reduces the training costs through three stages: training diffusion on downsampled data, distilling the pretrained diffusion, and progressive super-resolution. With the proposed pipeline, PaGoDA achieves a $64\times$ reduced cost in training its diffusion model on $8\times$ downsampled data; while at the inference, with the single-step, it performs state-of-the-art on ImageNet across all resolutions from $64\times64$ to $512\times512$, and text-to-image. PaGoDA's pipeline can be applied directly in the latent space, adding compression alongside the pre-trained autoencoder in Latent Diffusion Models (e.g., Stable Diffusion). The code is available at https://github.com/sony/pagoda.

Generating novel views from a single image remains a challenging task due to the complexity of 3D scenes and the limited diversity in the existing multi-view datasets to train a model on. Recent research combining large-scale text-to-image (T2I) models with monocular depth estimation (MDE) has shown promise in handling in-the-wild images. In these methods, an input view is geometrically warped to novel views with estimated depth maps, then the warped image is inpainted by T2I models. However, they struggle with noisy depth maps and loss of semantic details when warping an input view to novel viewpoints. In this paper, we propose a novel approach for single-shot novel view synthesis, a semantic-preserving generative warping framework that enables T2I generative models to learn where to warp and where to generate, through augmenting cross-view attention with self-attention. Our approach addresses the limitations of existing methods by conditioning the generative model on source view images and incorporating geometric warping signals. Qualitative and quantitative evaluations demonstrate that our model outperforms existing methods in both in-domain and out-of-domain scenarios. Project page is available at https://GenWarp-NVS.github.io.

We propose a new method for novelty detection that can tolerate high corruption of the training points, whereas previous works assumed either no or very low corruption. Our method trains a robust variational autoencoder (VAE), which aims to generate a model for the uncorrupted training points. To gain robustness to high corruption, we incorporate the following four changes to the common VAE: 1. Extracting crucial features of the latent code by a carefully designed dimension reduction component for distributions; 2. Modeling the latent distribution as a mixture of Gaussian low-rank inliers and full-rank outliers, where the testing only uses the inlier model; 3. Applying the Wasserstein-1 metric for regularization, instead of the Kullback-Leibler (KL) divergence; and 4. Using a robust error for reconstruction. We establish both robustness to outliers and suitability to low-rank modeling of the Wasserstein metric as opposed to the KL divergence. We illustrate state-of-the-art results on standard benchmarks.

Removing reverb from reverberant music is a necessary technique to clean up audio for downstream music manipulations. Reverberation of music contains two categories, natural reverb, and artificial reverb. Artificial reverb has a wider diversity than natural reverb due to its various parameter setups and reverberation types. However, recent supervised dereverberation methods may fail because they rely on sufficiently diverse and numerous pairs of reverberant observations and retrieved data for training in order to be generalizable to unseen observations during inference. To resolve these problems, we propose an unsupervised method that can remove a general kind of artificial reverb for music without requiring pairs of data for training. The proposed method is based on diffusion models, where it initializes the unknown reverberation operator with a conventional signal processing technique and simultaneously refines the estimate with the help of diffusion models. We show through objective and perceptual evaluations that our method outperforms the current leading vocal dereverberation benchmarks.

Score-based generative models (SGMs) learn a family of noise-conditional score functions corresponding to the data density perturbed with increasingly large amounts of noise. These perturbed data densities are linked together by the Fokker-Planck equation (FPE), a partial differential equation (PDE) governing the spatial-temporal evolution of a density undergoing a diffusion process. In this work, we derive a corresponding equation called the score FPE that characterizes the noise-conditional scores of the perturbed data densities (i.e., their gradients). Surprisingly, despite the impressive empirical performance, we observe that scores learned through denoising score matching (DSM) fail to fulfill the underlying score FPE, which is an inherent self-consistency property of the ground truth score. We prove that satisfying the score FPE is desirable as it improves the likelihood and the degree of conservativity. Hence, we propose to regularize the DSM objective to enforce satisfaction of the score FPE, and we show the effectiveness of this approach across various datasets.

Pre-trained diffusion models have been successfully used as priors in a variety of linear inverse problems, where the goal is to reconstruct a signal from noisy linear measurements. However, existing approaches require knowledge of the linear operator. In this paper, we propose GibbsDDRM, an extension of Denoising Diffusion Restoration Models (DDRM) to a blind setting in which the linear measurement operator is unknown. GibbsDDRM constructs a joint distribution of the data, measurements, and linear operator by using a pre-trained diffusion model for the data prior, and it solves the problem by posterior sampling with an efficient variant of a Gibbs sampler. The proposed method is problem-agnostic, meaning that a pre-trained diffusion model can be applied to various inverse problems without fine-tuning. In experiments, it achieved high performance on both blind image deblurring and vocal dereverberation tasks, despite the use of simple generic priors for the underlying linear operators.

One noted issue of vector-quantized variational autoencoder (VQ-VAE) is that the learned discrete representation uses only a fraction of the full capacity of the codebook, also known as codebook collapse. We hypothesize that the training scheme of VQ-VAE, which involves some carefully designed heuristics, underlies this issue. In this paper, we propose a new training scheme that extends the standard VAE via novel stochastic dequantization and quantization, called stochastically quantized variational autoencoder (SQ-VAE). In SQ-VAE, we observe a trend that the quantization is stochastic at the initial stage of the training but gradually converges toward a deterministic quantization, which we call self-annealing. Our experiments show that SQ-VAE improves codebook utilization without using common heuristics. Furthermore, we empirically show that SQ-VAE is superior to VAE and VQ-VAE in vision- and speech-related tasks.

Variational autoencoders (VAEs) often suffer from posterior collapse, which is a phenomenon in which the learned latent space becomes uninformative. This is often related to the hyperparameter resembling the data variance. It can be shown that an inappropriate choice of this hyperparameter causes the oversmoothness in the linearly approximated case and can be empirically verified for the general cases. Moreover, determining such appropriate choice becomes infeasible if the data variance is non-uniform or conditional. Therefore, we propose VAE extensions with generalized parameterizations of the data variance and incorporate maximum likelihood estimation into the objective function to adaptively regularize the decoder smoothness. The images generated from proposed VAE extensions show improved Fréchet inception distance (FID) on MNIST and CelebA datasets.