Text-to-audio systems, while increasingly performant, are slow at inference time, thus making their latency unpractical for many creative applications. We present Adversarial Relativistic-Contrastive (ARC) post-training, the first adversarial acceleration algorithm for diffusion/flow models not based on distillation. While past adversarial post-training methods have struggled to compare against their expensive distillation counterparts, ARC post-training is a simple procedure that (1) extends a recent relativistic adversarial formulation to diffusion/flow post-training and (2) combines it with a novel contrastive discriminator objective to encourage better prompt adherence. We pair ARC post-training with a number optimizations to Stable Audio Open and build a model capable of generating $\approx$12s of 44.1kHz stereo audio in $\approx$75ms on an H100, and $\approx$7s on a mobile edge-device, the fastest text-to-audio model to our knowledge.
Learning to associate audio with textual descriptions is valuable for a range of tasks, including pretraining, zero-shot classification, audio retrieval, audio captioning, and text-conditioned audio generation. Existing contrastive language-audio pretrained models are typically trained using global, clip-level descriptions, which provide only weak temporal supervision. We hypothesize that CLAP-like language-audio models - particularly, if they are expected to produce frame-level embeddings - can benefit from a stronger temporal supervision. To confirm our hypothesis, we curate a novel dataset of approximately 12,000 audio recordings from Freesound, each annotated with single-sentence free-text descriptions linked to a specific temporal segment in an audio recording. We use large language models to clean these annotations by removing references to non-audible events, transcribed speech, typos, and annotator language bias. We further propose a frame-wise contrastive training strategy that learns to align text descriptions with temporal regions in an audio recording and demonstrate that our model has better temporal text-audio alignment abilities compared to models trained only on global captions when evaluated on the AudioSet Strong benchmark. The dataset and our source code are available on Zenodo and GitHub, respectively.
In this work, we address the voice conversion (VC) task using a vector-based interface. To align audio embeddings between speakers, we employ discrete optimal transport mapping. Our evaluation results demonstrate the high quality and effectiveness of this method. Additionally, we show that applying discrete optimal transport as a post-processing step in audio generation can lead to the incorrect classification of synthetic audio as real.
Score-based Generative Models (SGMs) approximate a data distribution by perturbing it with Gaussian noise and subsequently denoising it via a learned reverse diffusion process. These models excel at modeling complex data distributions and generating diverse samples, achieving state-of-the-art performance across domains such as computer vision, audio generation, reinforcement learning, and computational biology. Despite their empirical success, existing Wasserstein-2 convergence analysis typically assume strong regularity conditions-such as smoothness or strict log-concavity of the data distribution-that are rarely satisfied in practice. In this work, we establish the first non-asymptotic Wasserstein-2 convergence guarantees for SGMs targeting semiconvex distributions with potentially discontinuous gradients. Our upper bounds are explicit and sharp in key parameters, achieving optimal dependence of $O(\sqrt{d})$ on the data dimension $d$ and convergence rate of order one. The framework accommodates a wide class of practically relevant distributions, including symmetric modified half-normal distributions, Gaussian mixtures, double-well potentials, and elastic net potentials. By leveraging semiconvexity without requiring smoothness assumptions on the potential such as differentiability, our results substantially broaden the theoretical foundations of SGMs, bridging the gap between empirical success and rigorous guarantees in non-smooth, complex data regimes.
Traditional video-to-audio generation techniques primarily focus on perspective video and non-spatial audio, often missing the spatial cues necessary for accurately representing sound sources in 3D environments. To address this limitation, we introduce a novel task, 360V2SA, to generate spatial audio from 360-degree videos, specifically producing First-order Ambisonics (FOA) audio - a standard format for representing 3D spatial audio that captures sound directionality and enables realistic 3D audio reproduction. We first create Sphere360, a novel dataset tailored for this task that is curated from real-world data. We also design an efficient semi-automated pipeline for collecting and cleaning paired video-audio data. To generate spatial audio from 360-degree video, we propose a novel framework OmniAudio, which leverages self-supervised pre-training using both spatial audio data (in FOA format) and large-scale non-spatial data. Furthermore, OmniAudio features a dual-branch framework that utilizes both panoramic and perspective video inputs to capture comprehensive local and global information from 360-degree videos. Experimental results demonstrate that OmniAudio achieves state-of-the-art performance across both objective and subjective metrics on Sphere360. Code and datasets are available at https://github.com/liuhuadai/OmniAudio. The project website is available at https://OmniAudio-360V2SA.github.io.
Recently, neural speech codecs (NSCs) trained as generative models have shown superior performance compared to conventional codecs at low bitrates. Although most state-of-the-art NSCs are trained as Generative Adversarial Networks (GANs), Diffusion Models (DMs), a recent class of generative models, represent a promising alternative due to their superior performance in image generation relative to GANs. Consequently, DMs have been successfully applied for audio and speech coding among various other audio generation applications. However, the design of diffusion-based NSCs has not yet been explored in a systematic way. We address this by providing a comprehensive analysis of diffusion-based NSCs divided into three contributions. First, we propose a categorization based on the conditioning and output domains of the DM. This simple conceptual framework allows us to define a design space for diffusion-based NSCs and to assign a category to existing approaches in the literature. Second, we systematically investigate unexplored designs by creating and evaluating new diffusion-based NSCs within the conceptual framework. Finally, we compare the proposed models to existing GAN and DM baselines through objective metrics and subjective listening tests.
The rapid advancement of audio generation technologies has escalated the risks of malicious deepfake audio across speech, sound, singing voice, and music, threatening multimedia security and trust. While existing countermeasures (CMs) perform well in single-type audio deepfake detection (ADD), their performance declines in cross-type scenarios. This paper is dedicated to studying the alltype ADD task. We are the first to comprehensively establish an all-type ADD benchmark to evaluate current CMs, incorporating cross-type deepfake detection across speech, sound, singing voice, and music. Then, we introduce the prompt tuning self-supervised learning (PT-SSL) training paradigm, which optimizes SSL frontend by learning specialized prompt tokens for ADD, requiring 458x fewer trainable parameters than fine-tuning (FT). Considering the auditory perception of different audio types,we propose the wavelet prompt tuning (WPT)-SSL method to capture type-invariant auditory deepfake information from the frequency domain without requiring additional training parameters, thereby enhancing performance over FT in the all-type ADD task. To achieve an universally CM, we utilize all types of deepfake audio for co-training. Experimental results demonstrate that WPT-XLSR-AASIST achieved the best performance, with an average EER of 3.58% across all evaluation sets. The code is available online.
Creating high-quality sound effects from videos and text prompts requires precise alignment between visual and audio domains, both semantically and temporally, along with step-by-step guidance for professional audio generation. However, current state-of-the-art video-guided audio generation models often fall short of producing high-quality audio for both general and specialized use cases. To address this challenge, we introduce a multi-stage, multi-modal, end-to-end generative framework with Chain-of-Thought-like (CoT-like) guidance learning, termed Chain-of-Perform (CoP). First, we employ a transformer-based network architecture designed to achieve CoP guidance, enabling the generation of both general and professional audio. Second, we implement a multi-stage training framework that follows step-by-step guidance to ensure the generation of high-quality sound effects. Third, we develop a CoP multi-modal dataset, guided by video, to support step-by-step sound effects generation. Evaluation results highlight the advantages of the proposed multi-stage CoP generative framework compared to the state-of-the-art models on a variety of datasets, with FAD 0.79 to 0.74 (+6.33%), CLIP 16.12 to 17.70 (+9.80%) on VGGSound, SI-SDR 1.98dB to 3.35dB (+69.19%), MOS 2.94 to 3.49(+18.71%) on PianoYT-2h, and SI-SDR 2.22dB to 3.21dB (+44.59%), MOS 3.07 to 3.42 (+11.40%) on Piano-10h.
Currently, high-quality, synchronized audio is synthesized using various multi-modal joint learning frameworks, leveraging video and optional text inputs. In the video-to-audio benchmarks, video-to-audio quality, semantic alignment, and audio-visual synchronization are effectively achieved. However, in real-world scenarios, speech and audio often coexist in videos simultaneously, and the end-to-end generation of synchronous speech and audio given video and text conditions are not well studied. Therefore, we propose an end-to-end multi-modal generation framework that simultaneously produces speech and audio based on video and text conditions. Furthermore, the advantages of video-to-audio (V2A) models for generating speech from videos remain unclear. The proposed framework, DeepAudio, consists of a video-to-audio (V2A) module, a text-to-speech (TTS) module, and a dynamic mixture of modality fusion (MoF) module. In the evaluation, the proposed end-to-end framework achieves state-of-the-art performance on the video-audio benchmark, video-speech benchmark, and text-speech benchmark. In detail, our framework achieves comparable results in the comparison with state-of-the-art models for the video-audio and text-speech benchmarks, and surpassing state-of-the-art models in the video-speech benchmark, with WER 16.57% to 3.15% (+80.99%), SPK-SIM 78.30% to 89.38% (+14.15%), EMO-SIM 66.24% to 75.56% (+14.07%), MCD 8.59 to 7.98 (+7.10%), MCD SL 11.05 to 9.40 (+14.93%) across a variety of dubbing settings.
Currently, high-quality, synchronized audio is synthesized from video and optional text inputs using various multi-modal joint learning frameworks. However, the precise alignment between the visual and generated audio domains remains far from satisfactory. One key factor is the lack of sufficient temporal and semantic alignment annotations in open-source video-audio and text-audio benchmarks. Therefore, we propose a framework for audio generation from videos, leveraging the internal chain-of-thought (CoT) of a multi-modal large language model (MLLM) to enable step-by-step reasoning without requiring additional annotations. Additionally, a corresponding multi-modal reasoning dataset is constructed to facilitate the learning of initial reasoning in audio generation. In the experiments, we demonstrate the effectiveness of the proposed framework in reducing misalignment (voice-over) in generated audio and achieving competitive performance compared to various state-of-the-art models. The evaluation results show that the proposed method outperforms state-of-the-art approaches across multiple metrics. Specifically, the F DP aSST indicator is reduced by up to 10.07%, the F DP AN N s indicator by up to 11.62%, and the F DV GG indicator by up to 38.61%. Furthermore, the IS indicator improves by up to 4.95%, the IB-score indicator increases by up to 6.39%, and the DeSync indicator is reduced by up to 0.89%.
Integrating audio comprehension and generation into large language models (LLMs) remains challenging due to the continuous nature of audio and the resulting high sampling rates. Here, we introduce a novel approach that combines Variational Quantization with Conditional Flow Matching to convert audio into ultra-low bitrate discrete tokens of 0.23kpbs, allowing for seamless integration with text tokens in LLMs. We fine-tuned a pretrained text-based LLM using Low-Rank Adaptation (LoRA) to assess its effectiveness in achieving true multimodal capabilities, i.e., audio comprehension and generation. Our tokenizer outperforms a traditional VQ-VAE across various datasets with diverse acoustic events. Despite the substantial loss of fine-grained details through audio tokenization, our multimodal LLM trained with discrete tokens achieves competitive results in audio comprehension with state-of-the-art methods, though audio generation is poor. Our results highlight the need for larger, more diverse datasets and improved evaluation metrics to advance multimodal LLM performance.
Recent works in cross-modal understanding and generation, notably through models like CLAP (Contrastive Language-Audio Pretraining) and CAVP (Contrastive Audio-Visual Pretraining), have significantly enhanced the alignment of text, video, and audio embeddings via a single contrastive loss. However, these methods often overlook the bidirectional interactions and inherent noises present in each modality, which can crucially impact the quality and efficacy of cross-modal integration. To address this limitation, we introduce DiffGAP, a novel approach incorporating a lightweight generative module within the contrastive space. Specifically, our DiffGAP employs a bidirectional diffusion process tailored to bridge the cross-modal gap more effectively. This involves a denoising process on text and video embeddings conditioned on audio embeddings and vice versa, thus facilitating a more nuanced and robust cross-modal interaction. Our experimental results on VGGSound and AudioCaps datasets demonstrate that DiffGAP significantly improves performance in video/text-audio generation and retrieval tasks, confirming its effectiveness in enhancing cross-modal understanding and generation capabilities.
As artificial intelligence-generated content (AIGC) continues to evolve, video-to-audio (V2A) generation has emerged as a key area with promising applications in multimedia editing, augmented reality, and automated content creation. While Transformer and Diffusion models have advanced audio generation, a significant challenge persists in extracting precise semantic information from videos, as current models often lose sequential context by relying solely on frame-based features. To address this, we present TA-V2A, a method that integrates language, audio, and video features to improve semantic representation in latent space. By incorporating large language models for enhanced video comprehension, our approach leverages text guidance to enrich semantic expression. Our diffusion model-based system utilizes automated text modulation to enhance inference quality and efficiency, providing personalized control through text-guided interfaces. This integration enhances semantic expression while ensuring temporal alignment, leading to more accurate and coherent video-to-audio generation.
Audio and music generation have emerged as crucial tasks in many applications, yet existing approaches face significant limitations: they operate in isolation without unified capabilities across modalities, suffer from scarce high-quality, multi-modal training data, and struggle to effectively integrate diverse inputs. In this work, we propose AudioX, a unified Diffusion Transformer model for Anything-to-Audio and Music Generation. Unlike previous domain-specific models, AudioX can generate both general audio and music with high quality, while offering flexible natural language control and seamless processing of various modalities including text, video, image, music, and audio. Its key innovation is a multi-modal masked training strategy that masks inputs across modalities and forces the model to learn from masked inputs, yielding robust and unified cross-modal representations. To address data scarcity, we curate two comprehensive datasets: vggsound-caps with 190K audio captions based on the VGGSound dataset, and V2M-caps with 6 million music captions derived from the V2M dataset. Extensive experiments demonstrate that AudioX not only matches or outperforms state-of-the-art specialized models, but also offers remarkable versatility in handling diverse input modalities and generation tasks within a unified architecture. The code and datasets will be available at https://zeyuet.github.io/AudioX/
Current audio generation conditioned by text or video focuses on aligning audio with text/video modalities. Despite excellent alignment results, these multimodal frameworks still cannot be directly applied to compelling movie storytelling involving multiple scenes, where "on-screen" sounds require temporally-aligned audio generation, while "off-screen" sounds contribute to appropriate environment sounds accompanied by background music when applicable. Inspired by professional movie production, this paper proposes a multi-agentic framework for audio generation supervised by an autonomous Sound Director agent, engaging multi-turn conversations with other agents for on-screen and off-screen sound generation through multimodal LLM. To address on-screen sound generation, after detecting any talking humans in videos, we capture semantically and temporally synchronized sound by training a prediction model that forecasts interpretable, time-varying audio control signals: loudness, pitch, and timbre, which are used by a Foley Artist agent to condition a cross-attention module in the sound generation. The Foley Artist works cooperatively with the Composer and Voice Actor agents, and together they autonomously generate off-screen sound to complement the overall production. Each agent takes on specific roles similar to those of a movie production team. To temporally ground audio language models, in ReelWave, text/video conditions are decomposed into atomic, specific sound generation instructions synchronized with visuals when applicable. Consequently, our framework can generate rich and relevant audio content conditioned on video clips extracted from movies.
Video-to-audio generation is essential for synthesizing realistic audio tracks that synchronize effectively with silent videos. Following the perspective of extracting essential signals from videos that can precisely control the mature text-to-audio generative diffusion models, this paper presents how to balance the representation of mel-spectrograms in terms of completeness and complexity through a new approach called Mel Quantization-Continuum Decomposition (Mel-QCD). We decompose the mel-spectrogram into three distinct types of signals, employing quantization or continuity to them, we can effectively predict them from video by a devised video-to-all (V2X) predictor. Then, the predicted signals are recomposed and fed into a ControlNet, along with a textual inversion design, to control the audio generation process. Our proposed Mel-QCD method demonstrates state-of-the-art performance across eight metrics, evaluating dimensions such as quality, synchronization, and semantic consistency. Our codes and demos will be released at \href{Website}{https://wjc2830.github.io/MelQCD/}.
Dynamic Magnetic Resonance Imaging (MRI) of the vocal tract has become an increasingly adopted imaging modality for speech motor studies. Beyond image signals, systematic data loss, noise pollution, and audio file corruption can occur due to the unpredictability of the MRI acquisition environment. In such cases, generating audio from images is critical for data recovery in both clinical and research applications. However, this remains challenging due to hardware constraints, acoustic interference, and data corruption. Existing solutions, such as denoising and multi-stage synthesis methods, face limitations in audio fidelity and generalizability. To address these challenges, we propose a Knowledge Enhanced Conditional Variational Autoencoder (KE-CVAE), a novel two-step "knowledge enhancement + variational inference" framework for generating speech audio signals from cine dynamic MRI sequences. This approach introduces two key innovations: (1) integration of unlabeled MRI data for knowledge enhancement, and (2) a variational inference architecture to improve generative modeling capacity. To the best of our knowledge, this is one of the first attempts at synthesizing speech audio directly from dynamic MRI video sequences. The proposed method was trained and evaluated on an open-source dynamic vocal tract MRI dataset recorded during speech. Experimental results demonstrate its effectiveness in generating natural speech waveforms while addressing MRI-specific acoustic challenges, outperforming conventional deep learning-based synthesis approaches.
Existing Existing automatic audio generation methods struggle to generate podcast-like audio programs effectively. The key challenges lie in in-depth content generation, appropriate and expressive voice production. This paper proposed PodAgent, a comprehensive framework for creating audio programs. PodAgent 1) generates informative topic-discussion content by designing a Host-Guest-Writer multi-agent collaboration system, 2) builds a voice pool for suitable voice-role matching and 3) utilizes LLM-enhanced speech synthesis method to generate expressive conversational speech. Given the absence of standardized evaluation criteria for podcast-like audio generation, we developed comprehensive assessment guidelines to effectively evaluate the model's performance. Experimental results demonstrate PodAgent's effectiveness, significantly surpassing direct GPT-4 generation in topic-discussion dialogue content, achieving an 87.4% voice-matching accuracy, and producing more expressive speech through LLM-guided synthesis. Demo page: https://podcast-agent.github.io/demo/. Source code: https://github.com/yujxx/PodAgent.
We introduce InspireMusic, a framework integrated super resolution and large language model for high-fidelity long-form music generation. A unified framework generates high-fidelity music, songs, and audio, which incorporates an autoregressive transformer with a super-resolution flow-matching model. This framework enables the controllable generation of high-fidelity long-form music at a higher sampling rate from both text and audio prompts. Our model differs from previous approaches, as we utilize an audio tokenizer with one codebook that contains richer semantic information, thereby reducing training costs and enhancing efficiency. This combination enables us to achieve high-quality audio generation with long-form coherence of up to $8$ minutes. Then, an autoregressive transformer model based on Qwen 2.5 predicts audio tokens. Next, we employ a super-resolution flow-matching model to generate high-sampling rate audio with fine-grained details learned from an acoustic codec model. Comprehensive experiments show that the InspireMusic-1.5B-Long model has a comparable performance to recent top-tier open-source systems, including MusicGen and Stable Audio 2.0, on subjective and objective evaluations. The code and pre-trained models are released at https://github.com/FunAudioLLM/InspireMusic.
Text-to-audio (TTA), which generates audio signals from textual descriptions, has received huge attention in recent years. However, recent works focused on text to monaural audio only. As we know, spatial audio provides more immersive auditory experience than monaural audio, e.g. in virtual reality. To address this issue, we propose a text-to-spatial-audio (TTSA) generation framework named DualSpec. Specifically, it first trains variational autoencoders (VAEs) for extracting the latent acoustic representations from sound event audio. Then, given text that describes sound events and event directions, the proposed method uses the encoder of a pretrained large language model to transform the text into text features. Finally, it trains a diffusion model from the latent acoustic representations and text features for the spatial audio generation. In the inference stage, only the text description is needed to generate spatial audio. Particularly, to improve the synthesis quality and azimuth accuracy of the spatial sound events simultaneously, we propose to use two kinds of acoustic features. One is the Mel spectrograms which is good for improving the synthesis quality, and the other is the short-time Fourier transform spectrograms which is good at improving the azimuth accuracy. We provide a pipeline of constructing spatial audio dataset with text prompts, for the training of the VAEs and diffusion model. We also introduce new spatial-aware evaluation metrics to quantify the azimuth errors of the generated spatial audio recordings. Experimental results demonstrate that the proposed method can generate spatial audio with high directional and event consistency.