Sang-gil Lee

Sang-gil Lee, Wei Ping, Boris Ginsburg et al.

Despite recent progress in generative adversarial network (GAN)-based vocoders, where the model generates raw waveform conditioned on acoustic features, it is challenging to synthesize high-fidelity audio for numerous speakers across various recording environments. In this work, we present BigVGAN, a universal vocoder that generalizes well for various out-of-distribution scenarios without fine-tuning. We introduce periodic activation function and anti-aliased representation into the GAN generator, which brings the desired inductive bias for audio synthesis and significantly improves audio quality. In addition, we train our GAN vocoder at the largest scale up to 112M parameters, which is unprecedented in the literature. We identify and address the failure modes in large-scale GAN training for audio, while maintaining high-fidelity output without over-regularization. Our BigVGAN, trained only on clean speech (LibriTTS), achieves the state-of-the-art performance for various zero-shot (out-of-distribution) conditions, including unseen speakers, languages, recording environments, singing voices, music, and instrumental audio. We release our code and model at: https://github.com/NVIDIA/BigVGAN

Sreyan Ghosh, Arushi Goel, Kaousheik Jayakumar et al.

We present Audio Flamingo Next (AF-Next), the next-generation and most capable large audio-language model in the Audio Flamingo series, designed to advance understanding and reasoning over speech, environmental sounds and music. Compared to Audio Flamingo 3, AF-Next introduces: (i) a stronger foundational audio-language model that significantly improves accuracy across diverse audio understanding tasks; (ii) scalable strategies for constructing large-scale audio understanding and reasoning data beyond existing academic benchmarks; (iii) support for long and complex audio inputs up to 30 minutes; and (iv) Temporal Audio Chain-of-Thought, a new reasoning paradigm that explicitly grounds intermediate reasoning steps to timestamps in long audio, enabling fine-grained temporal alignment and improved interpretability. To enable these capabilities, we first conduct a systematic analysis of Audio Flamingo 3 to identify key gaps in audio understanding and reasoning. We then curate and scale new large-scale datasets totaling over 1 million hours to address these limitations and expand the existing AudioSkills-XL, LongAudio-XL, AF-Think and AF-Chat datasets. AF-Next is trained using a curriculum-based strategy spanning pre-training, mid-training and post-training stages. Extensive experiments across 20 audio understanding and reasoning benchmarks, including challenging long-audio tasks, show that AF-Next outperforms similarly sized open models by large margins and remains highly competitive with and sometimes surpasses, much larger open-weight and closed models. Beyond benchmark performance, AF-Next exhibits strong real-world utility and transfers well to unseen tasks, highlighting its robustness and generalization ability. In addition to all data, code and methods, we open-source 3 variants of AF-Next, including AF-Next-Instruct, AF-Next-Think and AF-Next-Captioner.

Edresson Casanova, Ryan Langman, Paarth Neekhara et al. · nvidia

Large language models (LLMs) have significantly advanced audio processing through audio codecs that convert audio into discrete tokens, enabling the application of language modeling techniques to audio data. However, audio codecs often operate at high frame rates, resulting in slow training and inference, especially for autoregressive models. To address this challenge, we present the Low Frame-rate Speech Codec (LFSC): a neural audio codec that leverages finite scalar quantization and adversarial training with large speech language models to achieve high-quality audio compression with a 1.89 kbps bitrate and 21.5 frames per second. We demonstrate that our novel codec can make the inference of LLM-based text-to-speech models around three times faster while improving intelligibility and producing quality comparable to previous models.

Chaehun Shin, Heeseung Kim, Che Hyun Lee et al.

Despite the fact that text-to-video (TTV) model has recently achieved remarkable success, there have been few approaches on TTV for its extension to video editing. Motivated by approaches on TTV models adapting from diffusion-based text-to-image (TTI) models, we suggest the video editing framework given only a pretrained TTI model and a single <text, video> pair, which we term Edit-A-Video. The framework consists of two stages: (1) inflating the 2D model into the 3D model by appending temporal modules and tuning on the source video (2) inverting the source video into the noise and editing with target text prompt and attention map injection. Each stage enables the temporal modeling and preservation of semantic attributes of the source video. One of the key challenges for video editing include a background inconsistency problem, where the regions not included for the edit suffer from undesirable and inconsistent temporal alterations. To mitigate this issue, we also introduce a novel mask blending method, termed as sparse-causal blending (SC Blending). We improve previous mask blending methods to reflect the temporal consistency so that the area where the editing is applied exhibits smooth transition while also achieving spatio-temporal consistency of the unedited regions. We present extensive experimental results over various types of text and videos, and demonstrate the superiority of the proposed method compared to baselines in terms of background consistency, text alignment, and video editing quality.

Tingle Li, Siddharth Gururani, Kevin J. Shih et al.

Generative video-to-audio (V2A) models produce highly plausible soundtracks, but it remains unclear whether they capture the underlying physical processes. Existing evaluations emphasize perceptual realism and overlook physical correctness under controlled interventions. In this paper, we introduce FlatSounds, a benchmark that audits the physical reasoning of V2A models through: 1) controlled counterfactual pairs in which a single physical factor is varied, and 2) single-video pattern tests that probe internal consistency and directional trends. These settings test whether the generated audio correctly reflects specific physical properties and timings. Our evaluation of state-of-the-art models reveals a consistent trade-off: models rely more on text captions than the visual stream to infer physics and semantics. Captions generally improve physical and semantic accuracy, but paradoxically degrade temporal alignment. Our results highlight the need to move beyond audio quality toward learning physical processes directly from pixels. Finally, we find that our physics-based metrics correlate strongly with human preference tests on our own data. Project webpage: https://research.nvidia.com/labs/cosmos-lab/flatsounds/

Sreyan Ghosh, Arushi Goel, Lasha Koroshinadze et al.

We introduce Music Flamingo, a novel large audio-language model designed to advance music (including song) understanding in foundational audio models. While audio-language research has progressed rapidly, music remains challenging due to its dynamic, layered, and information-dense nature. Progress has been further limited by the difficulty of scaling open audio understanding models, primarily because of the scarcity of high-quality music data and annotations. As a result, prior models are restricted to producing short, high-level captions, answering only surface-level questions, and showing limited generalization across diverse musical cultures. To address these challenges, we curate MF-Skills, a large-scale dataset labeled through a multi-stage pipeline that yields rich captions and question-answer pairs covering harmony, structure, timbre, lyrics, and cultural context. We fine-tune an enhanced Audio Flamingo 3 backbone on MF-Skills and further strengthen multiple skills relevant to music understanding. To improve the model's reasoning abilities, we introduce a post-training recipe: we first cold-start with MF-Think, a novel chain-of-thought dataset grounded in music theory, followed by GRPO-based reinforcement learning with custom rewards. Music Flamingo achieves state-of-the-art results across 10+ benchmarks for music understanding and reasoning, establishing itself as a generalist and musically intelligent audio-language model. Beyond strong empirical results, Music Flamingo sets a new standard for advanced music understanding by demonstrating how models can move from surface-level recognition toward layered, human-like perception of songs. We believe this work provides both a benchmark and a foundation for the community to build the next generation of models that engage with music as meaningfully as humans do.

Arushi Goel, Sreyan Ghosh, Jaehyeon Kim et al.

We present Audio Flamingo 3 (AF3), a fully open state-of-the-art (SOTA) large audio-language model that advances reasoning and understanding across speech, sound, and music. AF3 introduces: (i) AF-Whisper, a unified audio encoder trained using a novel strategy for joint representation learning across all 3 modalities of speech, sound, and music; (ii) flexible, on-demand thinking, allowing the model to do chain-of-thought-type reasoning before answering; (iii) multi-turn, multi-audio chat; (iv) long audio understanding and reasoning (including speech) up to 10 minutes; and (v) voice-to-voice interaction. To enable these capabilities, we propose several large-scale training datasets curated using novel strategies, including AudioSkills-XL, LongAudio-XL, AF-Think, and AF-Chat, and train AF3 with a novel five-stage curriculum-based training strategy. Trained on only open-source audio data, AF3 achieves new SOTA results on over 20+ (long) audio understanding and reasoning benchmarks, surpassing both open-weight and closed-source models trained on much larger datasets.

Sang-gil Lee, Zhifeng Kong, Arushi Goel et al.

Recent years have seen significant progress in Text-To-Audio (TTA) synthesis, enabling users to enrich their creative workflows with synthetic audio generated from natural language prompts. Despite this progress, the effects of data, model architecture, training objective functions, and sampling strategies on target benchmarks are not well understood. With the purpose of providing a holistic understanding of the design space of TTA models, we set up a large-scale empirical experiment focused on diffusion and flow matching models. Our contributions include: 1) AF-Synthetic, a large dataset of high quality synthetic captions obtained from an audio understanding model; 2) a systematic comparison of different architectural, training, and inference design choices for TTA models; 3) an analysis of sampling methods and their Pareto curves with respect to generation quality and inference speed. We leverage the knowledge obtained from this extensive analysis to propose our best model dubbed Elucidated Text-To-Audio (ETTA). When evaluated on AudioCaps and MusicCaps, ETTA provides improvements over the baselines trained on publicly available data, while being competitive with models trained on proprietary data. Finally, we show ETTA's improved ability to generate creative audio following complex and imaginative captions -- a task that is more challenging than current benchmarks.

Zhifeng Kong, Kevin J Shih, Weili Nie et al.

Real-world audio is often degraded by numerous factors. This work presents an audio restoration model tailored for high-res music at 44.1kHz. Our model, Audio-to-Audio Schrödinger Bridges (A2SB), is capable of both bandwidth extension (predicting high-frequency components) and inpainting (re-generating missing segments). Critically, A2SB is end-to-end requiring no vocoder to predict waveform outputs, able to restore hour-long audio inputs, and trained on permissively licensed music data. A2SB is capable of achieving state-of-the-art band-width extension and inpainting quality on several out-of-distribution music test sets.

Alexander H. Liu, Sang-gil Lee, Chao-Han Huck Yang et al.

Pre-training and representation learning have been playing an increasingly important role in modern speech processing. Nevertheless, different applications have been relying on different foundation models, since predominant pre-training techniques are either designed for discriminative tasks or generative tasks. In this work, we make the first attempt at building a unified pre-training framework for both types of tasks in speech. We show that with the appropriate design choices for pre-training, one can jointly learn a representation encoder and generative audio decoder that can be applied to both types of tasks. We propose UniWav, an encoder-decoder framework designed to unify pre-training representation learning and generative tasks. On speech recognition, text-to-speech, and speech tokenization, UniWav achieves comparable performance to different existing foundation models, each trained on a specific task. Our findings suggest that a single general-purpose foundation model for speech can be built to replace different foundation models, reducing the overhead and cost of pre-training.

Jinchuan Tian, Sang-gil Lee, Zhifeng Kong et al. · nvidia

Recent advances in the audio language modeling (ALM) domain tackle audio understanding and text-to-audio generation as separate tasks. Very few studies attempt to unify these tasks -- an essential step toward advanced multimodal reasoning. This paper introduces U}nified Audio Language Model (UALM), which aims to unify audio understanding, text-to-audio generation, and multimodal reasoning in a single model. To achieve this goal, we first present UALM-Gen, a text-to-audio language model that directly predicts audio tokens and is comparable to state-of-the-art diffusion-based models. We then demonstrate, using proper data blending, training recipes, and inference techniques, that our single UALM model matches the quality of state-of-the-art specialized models in audio understanding, text-to-audio generation, and text reasoning. Furthermore, we present UALM-Reason, a multimodal reasoning model that utilizes both text and audio in the intermediate thinking steps to facilitate complex generation tasks. To our knowledge, this is the first demonstration in audio research of cross-modal generative reasoning, with its effectiveness confirmed by subjective evaluations.

Zhifeng Kong, Sang-gil Lee, Deepanway Ghosal et al.

It is an open challenge to obtain high quality training data, especially captions, for text-to-audio models. Although prior methods have leveraged \textit{text-only language models} to augment and improve captions, such methods have limitations related to scale and coherence between audio and captions. In this work, we propose an audio captioning pipeline that uses an \textit{audio language model} to synthesize accurate and diverse captions for audio at scale. We leverage this pipeline to produce a dataset of synthetic captions for AudioSet, named \texttt{AF-AudioSet}, and then evaluate the benefit of pre-training text-to-audio models on these synthetic captions. Through systematic evaluations on AudioCaps and MusicCaps, we find leveraging our pipeline and synthetic captions leads to significant improvements on audio generation quality, achieving a new \textit{state-of-the-art}.

Sang-gil Lee, Eunji Kim, Jae Seok Bae et al.

The computer-aided diagnosis of focal liver lesions (FLLs) can help improve workflow and enable correct diagnoses; FLL detection is the first step in such a computer-aided diagnosis. Despite the recent success of deep-learning-based approaches in detecting FLLs, current methods are not sufficiently robust for assessing misaligned multiphase data. By introducing an attention-guided multiphase alignment in feature space, this study presents a fully automated, end-to-end learning framework for detecting FLLs from multiphase computed tomography (CT) images. Our method is robust to misaligned multiphase images owing to its complete learning-based approach, which reduces the sensitivity of the model's performance to the quality of registration and enables a standalone deployment of the model in clinical practice. Evaluation on a large-scale dataset with 280 patients confirmed that our method outperformed previous state-of-the-art methods and significantly reduced the performance degradation for detecting FLLs using misaligned multiphase CT images. The robustness of the proposed method can enhance the clinical adoption of the deep-learning-based computer-aided detection system.

Sang-gil Lee, Heeseung Kim, Chaehun Shin et al.

Denoising diffusion probabilistic models have been recently proposed to generate high-quality samples by estimating the gradient of the data density. The framework defines the prior noise as a standard Gaussian distribution, whereas the corresponding data distribution may be more complicated than the standard Gaussian distribution, which potentially introduces inefficiency in denoising the prior noise into the data sample because of the discrepancy between the data and the prior. In this paper, we propose PriorGrad to improve the efficiency of the conditional diffusion model for speech synthesis (for example, a vocoder using a mel-spectrogram as the condition) by applying an adaptive prior derived from the data statistics based on the conditional information. We formulate the training and sampling procedures of PriorGrad and demonstrate the advantages of an adaptive prior through a theoretical analysis. Focusing on the speech synthesis domain, we consider the recently proposed diffusion-based speech generative models based on both the spectral and time domains and show that PriorGrad achieves faster convergence and inference with superior performance, leading to an improved perceptual quality and robustness to a smaller network capacity, and thereby demonstrating the efficiency of a data-dependent adaptive prior.

Sang-gil Lee, Sungwon Kim, Sungroh Yoon

Normalizing flows (NFs) have become a prominent method for deep generative models that allow for an analytic probability density estimation and efficient synthesis. However, a flow-based network is considered to be inefficient in parameter complexity because of reduced expressiveness of bijective mapping, which renders the models unfeasibly expensive in terms of parameters. We present an alternative parameterization scheme called NanoFlow, which uses a single neural density estimator to model multiple transformation stages. Hence, we propose an efficient parameter decomposition method and the concept of flow indication embedding, which are key missing components that enable density estimation from a single neural network. Experiments performed on audio and image models confirm that our method provides a new parameter-efficient solution for scalable NFs with significant sublinear parameter complexity.

Sungwon Kim, Sang-gil Lee, Jongyoon Song et al.

Most modern text-to-speech architectures use a WaveNet vocoder for synthesizing high-fidelity waveform audio, but there have been limitations, such as high inference time, in its practical application due to its ancestral sampling scheme. The recently suggested Parallel WaveNet and ClariNet have achieved real-time audio synthesis capability by incorporating inverse autoregressive flow for parallel sampling. However, these approaches require a two-stage training pipeline with a well-trained teacher network and can only produce natural sound by using probability distillation along with auxiliary loss terms. We propose FloWaveNet, a flow-based generative model for raw audio synthesis. FloWaveNet requires only a single-stage training procedure and a single maximum likelihood loss, without any additional auxiliary terms, and it is inherently parallel due to the characteristics of generative flow. The model can efficiently sample raw audio in real-time, with clarity comparable to previous two-stage parallel models. The code and samples for all models, including our FloWaveNet, are publicly available.

Sang-gil Lee, Jae Seok Bae, Hyunjae Kim et al.

We present a focal liver lesion detection model leveraged by custom-designed multi-phase computed tomography (CT) volumes, which reflects real-world clinical lesion detection practice using a Single Shot MultiBox Detector (SSD). We show that grouped convolutions effectively harness richer information of the multi-phase data for the object detection model, while a naive application of SSD suffers from a generalization gap. We trained and evaluated the modified SSD model and recently proposed variants with our CT dataset of 64 subjects by five-fold cross validation. Our model achieved a 53.3% average precision score and ran in under three seconds per volume, outperforming the original model and state-of-the-art variants. Results show that the one-stage object detection model is a practical solution, which runs in near real-time and can learn an unbiased feature representation from a large-volume real-world detection dataset, which requires less tedious and time consuming construction of the weak phase-level bounding box labels.

Sang-gil Lee, Uiwon Hwang, Seonwoo Min et al.

We propose an application of sequence generative adversarial networks (SeqGAN), which are generative adversarial networks for discrete sequence generation, for creating polyphonic musical sequences. Instead of a monophonic melody generation suggested in the original work, we present an efficient representation of a polyphony MIDI file that simultaneously captures chords and melodies with dynamic timings. The proposed method condenses duration, octaves, and keys of both melodies and chords into a single word vector representation, and recurrent neural networks learn to predict distributions of sequences from the embedded musical word space. We experiment with the original method and the least squares method to the discriminator, which is known to stabilize the training of GANs. The network can create sequences that are musically coherent and shows an improved quantitative and qualitative measures. We also report that careful optimization of reinforcement learning signals of the model is crucial for general application of the model.

Seongsik Park, Sang-gil Lee, Hyunha Nam et al.

Nowadays deep learning is dominating the field of machine learning with state-of-the-art performance in various application areas. Recently, spiking neural networks (SNNs) have been attracting a great deal of attention, notably owning to their power efficiency, which can potentially allow us to implement a low-power deep learning engine suitable for real-time/mobile applications. However, implementing SNN-based deep learning remains challenging, especially gradient-based training of SNNs by error backpropagation. We cannot simply propagate errors through SNNs in conventional way because of the property of SNNs that process discrete data in the form of a series. Consequently, most of the previous studies employ a workaround technique, which first trains a conventional weighted-sum deep neural network and then maps the learning weights to the SNN under training, instead of training SNN parameters directly. In order to eliminate this workaround, recently proposed is a new class of SNN named deep spiking networks (DSNs), which can be trained directly (without a mapping from conventional deep networks) by error backpropagation with stochastic gradient descent. In this paper, we show that the initialization of the membrane potential on the backward path is an important step in DSN training, through diverse experiments performed under various conditions. Furthermore, we propose a simple and efficient method that can improve DSN training by controlling the initial membrane potential on the backward path. In our experiments, adopting the proposed approach allowed us to boost the performance of DSN training in terms of converging time and accuracy.