Koichi Saito

Naoki Murata, Koichi Saito, Chieh-Hsin Lai et al.

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.

Koichi Saito, Naoki Murata, Toshimitsu Uesaka et al.

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.

Yin-Jyun Luo, Kin Wai Cheuk, Woosung Choi et al.

Existing work on pitch and timbre disentanglement has been mostly focused on single-instrument music audio, excluding the cases where multiple instruments are presented. To fill the gap, we propose DisMix, a generative framework in which the pitch and timbre representations act as modular building blocks for constructing the melody and instrument of a source, and the collection of which forms a set of per-instrument latent representations underlying the observed mixture. By manipulating the representations, our model samples mixtures with novel combinations of pitch and timbre of the constituent instruments. We can jointly learn the disentangled pitch-timbre representations and a latent diffusion transformer that reconstructs the mixture conditioned on the set of source-level representations. We evaluate the model using both a simple dataset of isolated chords and a realistic four-part chorales in the style of J.S. Bach, identify the key components for the success of disentanglement, and demonstrate the application of mixture transformation based on source-level attribute manipulation.

Christian Simon, Masato Ishii, Wei-Yao Wang et al.

Scaling multimodal alignment between video and audio is challenging, particularly due to limited data and the mismatch between text descriptions and frame-level video information. In this work, we tackle the scaling challenge in multimodal-to-audio generation, examining whether models trained on short instances can generalize to longer ones during testing. To tackle this challenge, we present multimodal hierarchical networks so-called MMHNet, an enhanced extension of state-of-the-art video-to-audio models. Our approach integrates a hierarchical method and non-causal Mamba to support long-form audio generation. Our proposed method significantly improves long audio generation up to more than 5 minutes. We also prove that training short and testing long is possible in the video-to-audio generation tasks without training on the longer durations. We show in our experiments that our proposed method could achieve remarkable results on long-video to audio benchmarks, beating prior works in video-to-audio tasks. Moreover, we showcase our model capability in generating more than 5 minutes, while prior video-to-audio methods fall short in generating with long durations.

Yichen Huang, Zachary Novack, Koichi Saito et al.

Despite significant recent advances in generative acoustic text-to-music (TTM) modeling, robust evaluation of these models lags behind, relying in particular on the popular Fréchet Audio Distance (FAD). In this work, we rigorously study the design space of reference-based divergence metrics for evaluating TTM models through (1) designing four synthetic meta-evaluations to measure sensitivity to particular musical desiderata, and (2) collecting and evaluating on MusicPrefs, the first open-source dataset of human preferences for TTM systems. We find that not only is the standard FAD setup inconsistent on both synthetic and human preference data, but that nearly all existing metrics fail to effectively capture desiderata, and are only weakly correlated with human perception. We propose a new metric, the MAUVE Audio Divergence (MAD), computed on representations from a self-supervised audio embedding model. We find that this metric effectively captures diverse musical desiderata (average rank correlation 0.84 for MAD vs. 0.49 for FAD and also correlates more strongly with MusicPrefs (0.62 vs. 0.14).

Yonghyun Kim, Wayne Chi, Anastasios N. Angelopoulos et al. · gatech

We present Music Arena, an open platform for scalable human preference evaluation of text-to-music (TTM) models. Soliciting human preferences via listening studies is the gold standard for evaluation in TTM, but these studies are expensive to conduct and difficult to compare, as study protocols may differ across systems. Moreover, human preferences might help researchers align their TTM systems or improve automatic evaluation metrics, but an open and renewable source of preferences does not currently exist. We aim to fill these gaps by offering *live* evaluation for TTM. In Music Arena, real-world users input text prompts of their choosing and compare outputs from two TTM systems, and their preferences are used to compile a leaderboard. While Music Arena follows recent evaluation trends in other AI domains, we also design it with key features tailored to music: an LLM-based routing system to navigate the heterogeneous type signatures of TTM systems, and the collection of *detailed* preferences including listening data and natural language feedback. We also propose a rolling data release policy with user privacy guarantees, providing a renewable source of preference data and increasing platform transparency. Through its standardized evaluation protocol, transparent data access policies, and music-specific features, Music Arena not only addresses key challenges in the TTM ecosystem but also demonstrates how live evaluation can be thoughtfully adapted to unique characteristics of specific AI domains. Music Arena is available at: https://music-arena.org . Preference data is available at: https://huggingface.co/music-arena .

Julian Tanke, Takashi Shibuya, Kengo Uchida et al.

Generating realistic dyadic human motion from text descriptions presents significant challenges, particularly for extended interactions that exceed typical training sequence lengths. While recent transformer-based approaches have shown promising results for short-term dyadic motion synthesis, they struggle with longer sequences due to inherent limitations in positional encoding schemes. In this paper, we introduce Dyadic Mamba, a novel approach that leverages State-Space Models (SSMs) to generate high-quality dyadic human motion of arbitrary length. Our method employs a simple yet effective architecture that facilitates information flow between individual motion sequences through concatenation, eliminating the need for complex cross-attention mechanisms. We demonstrate that Dyadic Mamba achieves competitive performance on standard short-term benchmarks while significantly outperforming transformer-based approaches on longer sequences. Additionally, we propose a new benchmark for evaluating long-term motion synthesis quality, providing a standardized framework for future research. Our results demonstrate that SSM-based architectures offer a promising direction for addressing the challenging task of long-term dyadic human motion synthesis from text descriptions.

Weihan Xu, Kan Jen Cheng, Koichi Saito et al.

Joint editing of audio and visual content is crucial for precise and controllable content creation. This new task poses challenges due to the limitations of paired audio-visual data before and after targeted edits, and the heterogeneity across modalities. To address the data and modeling challenges in joint audio-visual editing, we introduce SAVEBench, a paired audiovisual dataset with text and mask conditions to enable object-grounded source-to-target learning. With SAVEBench, we train the Schrodinger Audio-Visual Editor (SAVE), an end-to-end flow-matching model that edits audio and video in parallel while keeping them aligned throughout processing. SAVE incorporates a Schrodinger Bridge that learns a direct transport from source to target audiovisual mixtures. Our evaluation demonstrates that the proposed SAVE model is able to remove the target objects in audio and visual content while preserving the remaining content, with stronger temporal synchronization and audiovisual semantic correspondence compared with pairwise combinations of an audio editor and a video editor.

Satvik Dixit, Koichi Saito, Zhi Zhong et al. · cmu

Video-to-audio generation (V2A) is of increasing importance in domains such as film post-production, AR/VR, and sound design, particularly for the creation of Foley sound effects synchronized with on-screen actions. Foley requires generating audio that is both semantically aligned with visible events and temporally aligned with their timing. Yet, there is a mismatch between evaluation and downstream applications due to the absence of a benchmark tailored to Foley-style scenarios. We find that 74% of videos from past evaluation datasets have poor audio-visual correspondence. Moreover, they are dominated by speech and music, domains that lie outside the use case for Foley. To address this gap, we introduce FoleyBench, the first large-scale benchmark explicitly designed for Foley-style V2A evaluation. FoleyBench contains 5,000 (video, ground-truth audio, text caption) triplets, each featuring visible sound sources with audio causally tied to on-screen events. The dataset is built using an automated, scalable pipeline applied to in-the-wild internet videos from YouTube-based and Vimeo-based sources. Compared to past datasets, we show that videos from FoleyBench have stronger coverage of sound categories from a taxonomy specifically designed for Foley sound. Each clip is further labeled with metadata capturing source complexity, UCS/AudioSet category, and video length, enabling fine-grained analysis of model performance and failure modes. We benchmark several state-of-the-art V2A models, evaluating them on audio quality, audio-video alignment, temporal synchronization, and audio-text consistency. Samples are available at: https://gclef-cmu.org/foleybench

Jiaben Chen, Zixin Wang, Ailing Zeng et al.

In this work, we present TalkCuts, a large-scale dataset designed to facilitate the study of multi-shot human speech video generation. Unlike existing datasets that focus on single-shot, static viewpoints, TalkCuts offers 164k clips totaling over 500 hours of high-quality human speech videos with diverse camera shots, including close-up, half-body, and full-body views. The dataset includes detailed textual descriptions, 2D keypoints and 3D SMPL-X motion annotations, covering over 10k identities, enabling multimodal learning and evaluation. As a first attempt to showcase the value of the dataset, we present Orator, an LLM-guided multi-modal generation framework as a simple baseline, where the language model functions as a multi-faceted director, orchestrating detailed specifications for camera transitions, speaker gesticulations, and vocal modulation. This architecture enables the synthesis of coherent long-form videos through our integrated multi-modal video generation module. Extensive experiments in both pose-guided and audio-driven settings show that training on TalkCuts significantly enhances the cinematographic coherence and visual appeal of generated multi-shot speech videos. We believe TalkCuts provides a strong foundation for future work in controllable, multi-shot speech video generation and broader multimodal learning.

Koichi Saito, Julian Tanke, Christian Simon et al.

Prevailing Video-to-Audio (V2A) generation models operate offline, assuming an entire video sequence or chunks of frames are available beforehand. This critically limits their use in interactive applications such as live content creation and emerging generative world models. To address this gap, we introduce the novel task of frame-level online V2A generation, where a model autoregressively generates audio from video without access to future video frames. Furthermore, we propose SoundReactor, which, to the best of our knowledge, is the first simple yet effective framework explicitly tailored for this task. Our design enforces end-to-end causality and targets low per-frame latency with audio-visual synchronization. Our model's backbone is a decoder-only causal transformer over continuous audio latents. For vision conditioning, it leverages grid (patch) features extracted from the smallest variant of the DINOv2 vision encoder, which are aggregated into a single token per frame to maintain end-to-end causality and efficiency. The model is trained through a diffusion pre-training followed by consistency fine-tuning to accelerate the diffusion head decoding. On a benchmark of diverse gameplay videos from AAA titles, our model successfully generates semantically and temporally aligned, high-quality full-band stereo audio, validated by both objective and human evaluations. Furthermore, our model achieves low per-frame waveform-level latency (26.3ms with the head NFE=1, 31.5ms with NFE=4) on 30FPS, 480p videos using a single H100. Demo samples are available at https://koichi-saito-sony.github.io/soundreactor/.

Koichi Saito, Stefan Uhlich, Giorgio Fabbro et al.

Recently, deep neural network (DNN)-based speech enhancement (SE) systems have been used with great success. During training, such systems require clean speech data - ideally, in large quantity with a variety of acoustic conditions, many different speaker characteristics and for a given sampling rate (e.g., 48kHz for fullband SE). However, obtaining such clean speech data is not straightforward - especially, if only considering publicly available datasets. At the same time, a lot of material for automatic speech recognition (ASR) with the desired acoustic/speaker/sampling rate characteristics is publicly available except being clean, i.e., it also contains background noise as this is even often desired in order to have ASR systems that are noise-robust. Hence, using such data to train SE systems is not straightforward. In this paper, we propose two improvements to train SE systems on noisy speech data. First, we propose several modifications of the loss functions, which make them robust against noisy speech targets. In particular, computing the median over the sample axis before averaging over time-frequency bins allows to use such data. Furthermore, we propose a noise augmentation scheme for mixture-invariant training (MixIT), which allows using it also in such scenarios. For our experiments, we use the Mozilla Common Voice dataset and we show that using our robust loss function improves PESQ by up to 0.19 compared to a system trained in the traditional way. Similarly, for MixIT we can see an improvement of up to 0.27 in PESQ when using our proposed noise augmentation.

Koichi Saito, Tomohiko Nakamura, Kohei Yatabe et al.

Audio source separation is often used as preprocessing of various applications, and one of its ultimate goals is to construct a single versatile model capable of dealing with the varieties of audio signals. Since sampling frequency, one of the audio signal varieties, is usually application specific, the preceding audio source separation model should be able to deal with audio signals of all sampling frequencies specified in the target applications. However, conventional models based on deep neural networks (DNNs) are trained only at the sampling frequency specified by the training data, and there are no guarantees that they work with unseen sampling frequencies. In this paper, we propose a convolution layer capable of handling arbitrary sampling frequencies by a single DNN. Through music source separation experiments, we show that the introduction of the proposed layer enables a conventional audio source separation model to consistently work with even unseen sampling frequencies.