Videos contain a wealth of information, and generating detailed and accurate descriptions in natural language is a key aspect of video understanding. In this paper, we present video-SALMONN 2, an advanced audio-visual large language model (LLM) with low-rank adaptation (LoRA) designed for enhanced video (with paired audio) captioning through directed preference optimisation (DPO). We propose new metrics to evaluate the completeness and accuracy of video descriptions, which are optimised using DPO. To further improve training, we propose a novel multi-round DPO (MrDPO) approach, which involves periodically updating the DPO reference model, merging and re-initialising the LoRA module as a proxy for parameter updates after each training round (1,000 steps), and incorporating guidance from ground-truth video captions to stabilise the process. Experimental results show that MrDPO significantly enhances video-SALMONN 2's captioning accuracy, reducing the captioning error rates by 28\%. The final video-SALMONN 2 model, with just 7 billion parameters, surpasses leading models such as GPT-4o and Gemini-1.5-Pro in video captioning tasks, while maintaining highly competitive performance to the state-of-the-art on widely used video question-answering benchmarks among models of similar size. Codes are available at \href{https://github.com/bytedance/video-SALMONN-2}{https://github.com/bytedance/video-SALMONN-2}.
We propose an instruction-following audio comprehension model that leverages the dialogue continuation ability of large language models (LLMs). Instead of directly generating target captions in training data, the proposed method trains a model to produce responses as if the input caption triggered a dialogue. This dialogue continuation training mitigates the caption variation problem. Learning to continue a dialogue effectively captures the caption's meaning beyond its surface-level words. As a result, our model enables zero-shot instruction-following capability without multitask instruction tuning, even trained solely on audio captioning datasets. Experiments on AudioCaps, WavCaps, and Clotho datasets with AudioBench audio-scene question-answering tests demonstrate our model's ability to follow various unseen instructions.
High-quality, large-scale audio captioning is crucial for advancing audio understanding, yet current automated methods often generate captions that lack fine-grained detail and contextual accuracy, primarily due to their reliance on limited unimodal or superficial multimodal information. Drawing inspiration from human auditory perception, which adeptly integrates cross-modal cues and performs sophisticated auditory scene analysis, we introduce a novel two-stage automated pipeline. This pipeline first employs specialized pretrained models to extract diverse contextual cues (e.g., speech, music, general sounds, and visual information from associated video). A large language model (LLM) then synthesizes these rich, multimodal inputs to generate detailed and context-aware audio captions. Key contributions of this work include: (1) the proposed scalable method for fine-grained audio caption generation; (2) FusionAudio, a new large-scale dataset comprising 1.2 million such detailed captions, combined with 6 million QA pairs; and (3) enhanced audio models developed using FusionAudio, specifically a CLAP-based audio encoder with superior audio-text alignment and instruction following. This paper paves the way for more nuanced and accurate automated understanding of complex audio environments. Code and data can be found in https://github.com/satsuki2486441738/FusionAudio.
Automated Audio Captioning (AAC) aims to describe the semantic contexts of general sounds, including acoustic events and scenes, by leveraging effective acoustic features. To enhance performance, an AAC method, EnCLAP, employed discrete tokens from EnCodec as an effective input for fine-tuning a language model BART. However, EnCodec is designed to reconstruct waveforms rather than capture the semantic contexts of general sounds, which AAC should describe. To address this issue, we propose CLAP-ART, an AAC method that utilizes ``semantic-rich and discrete'' tokens as input. CLAP-ART computes semantic-rich discrete tokens from pre-trained audio representations through vector quantization. We experimentally confirmed that CLAP-ART outperforms baseline EnCLAP on two AAC benchmarks, indicating that semantic-rich discrete tokens derived from semantically rich AR are beneficial for AAC.
Current vision-guided audio captioning systems frequently fail to address audiovisual misalignment in real-world scenarios, such as dubbed content or off-screen sounds. To bridge this critical gap, we present an entropy-aware gated fusion framework that dynamically modulates visual information flow through cross-modal uncertainty quantification. Our novel approach employs attention entropy analysis in cross-attention layers to automatically identify and suppress misleading visual cues during modal fusion. Complementing this architecture, we develop a batch-wise audiovisual shuffling technique that generates synthetic mismatched training pairs, greatly enhancing model resilience against alignment noise. Evaluations on the AudioCaps benchmark demonstrate our system's superior performance over existing baselines, especially in mismatched modality scenarios. Furthermore, our solution demonstrates an approximately 6x improvement in inference speed compared to the baseline.
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.
Contrastive language-audio pre-training (CLAP) has addressed audio-language tasks such as audio-text retrieval by aligning audio and text in a common feature space. While CLAP addresses general audio-language tasks, its audio features do not generalize well in audio tasks. In contrast, self-supervised learning (SSL) models learn general-purpose audio features that perform well in diverse audio tasks. We pursue representation learning that can be widely used in audio applications and hypothesize that a method that learns both general audio features and CLAP features should achieve our goal, which we call a general-purpose audio-language representation. To implement our hypothesis, we propose M2D2, a second-generation masked modeling duo (M2D) that combines an SSL M2D and CLAP. M2D2 learns two types of features using two modalities (audio and text) in a two-stage training process. It also utilizes advanced LLM-based sentence embeddings in CLAP training for powerful semantic supervision. In the first stage, M2D2 learns generalizable audio features from M2D and CLAP, where CLAP aligns the features with the fine LLM-based semantic embeddings. In the second stage, it learns CLAP features using the audio features learned from the LLM-based embeddings. Through these pre-training stages, M2D2 should enhance generalizability and performance in its audio and CLAP features. Experiments validated that M2D2 achieves effective general-purpose audio-language representation, highlighted with SOTA fine-tuning mAP of 49.0 for AudioSet, SOTA performance in music tasks, and top-level performance in audio-language tasks.
Large Language Models (LLMs) have recently shown remarkable ability to process not only text but also multimodal inputs such as speech and audio. However, most existing models primarily focus on analyzing input signals using text instructions, overlooking scenarios in which speech instructions and audio are mixed and serve as inputs to the model. To address these challenges, we introduce Solla, a novel framework designed to understand speech-based questions and hear the acoustic context concurrently. Solla incorporates an audio tagging module to effectively identify and represent audio events, as well as an ASR-assisted prediction method to improve comprehension of spoken content. To rigorously evaluate Solla and other publicly available models, we propose a new benchmark dataset called SA-Eval, which includes three tasks: audio event classification, audio captioning, and audio question answering. SA-Eval has diverse speech instruction with various speaking styles, encompassing two difficulty levels, easy and hard, to capture the range of real-world acoustic conditions. Experimental results show that Solla performs on par with or outperforms baseline models on both the easy and hard test sets, underscoring its effectiveness in jointly understanding speech and audio.
Multimodal Audio-Language Models (ALMs) can understand and reason over both audio and text. Typically, reasoning performance correlates with model size, with the best results achieved by models exceeding 8 billion parameters. However, no prior work has explored enabling small audio-language models to perform reasoning tasks, despite the potential applications for edge devices. To address this gap, we introduce Mellow, a small Audio-Language Model specifically designed for reasoning. Mellow achieves state-of-the-art performance among existing small audio-language models and surpasses several larger models in reasoning capabilities. For instance, Mellow scores 52.11 on MMAU, comparable to SoTA Qwen2 Audio (which scores 52.5) while using 50 times fewer parameters and being trained on 60 times less data (audio hrs). To train Mellow, we introduce ReasonAQA, a dataset designed to enhance audio-grounded reasoning in models. It consists of a mixture of existing datasets (30% of the data) and synthetically generated data (70%). The synthetic dataset is derived from audio captioning datasets, where Large Language Models (LLMs) generate detailed and multiple-choice questions focusing on audio events, objects, acoustic scenes, signal properties, semantics, and listener emotions. To evaluate Mellow's reasoning ability, we benchmark it on a diverse set of tasks, assessing on both in-distribution and out-of-distribution data, including audio understanding, deductive reasoning, and comparative reasoning. Finally, we conduct extensive ablation studies to explore the impact of projection layer choices, synthetic data generation methods, and language model pretraining on reasoning performance. Our training dataset, findings, and baseline pave the way for developing small ALMs capable of reasoning.
Understanding and reasoning over non-speech sounds and music are crucial for both humans and AI agents to interact effectively with their environments. In this paper, we introduce Audio Flamingo 2 (AF2), an Audio-Language Model (ALM) with advanced audio understanding and reasoning capabilities. AF2 leverages (i) a custom CLAP model, (ii) synthetic Audio QA data for fine-grained audio reasoning, and (iii) a multi-stage curriculum learning strategy. AF2 achieves state-of-the-art performance with only a 3B parameter small language model, surpassing large open-source and proprietary models across over 20 benchmarks. Next, for the first time, we extend audio understanding to long audio segments (30 secs to 5 mins) and propose LongAudio, a large and novel dataset for training ALMs on long audio captioning and question-answering tasks. Fine-tuning AF2 on LongAudio leads to exceptional performance on our proposed LongAudioBench, an expert annotated benchmark for evaluating ALMs on long audio understanding capabilities. We conduct extensive ablation studies to confirm the efficacy of our approach. Project Website: https://research.nvidia.com/labs/adlr/AF2/.
Connecting audio encoders with large language models (LLMs) allows the LLM to perform various audio understanding tasks, such as automatic speech recognition (ASR) and audio captioning (AC). Most research focuses on training an adapter layer to generate a unified audio feature for the LLM. However, different tasks may require distinct features that emphasize either semantic or acoustic aspects, making task-specific audio features more desirable. In this paper, we propose Prompt-aware Mixture (PaM) to enhance the Speech LLM that uses multiple audio encoders. Our approach involves using different experts to extract different features based on the prompt that indicates different tasks. Experiments demonstrate that with PaM, only one Speech LLM surpasses the best performances achieved by all single-encoder Speech LLMs on ASR, Speaker Number Verification, and AC tasks. PaM also outperforms other feature fusion baselines, such as concatenation and averaging.
Teacher-forcing training for audio captioning usually leads to exposure bias due to training and inference mismatch. Prior works propose the contrastive method to deal with caption degeneration. However, the contrastive method ignores the temporal information when measuring similarity across acoustic and linguistic modalities, leading to inferior performance. In this work, we develop the temporal-similarity score by introducing the unbiased sliced Wasserstein RBF (USW-RBF) kernel equipped with rotary positional embedding to account for temporal information across modalities. In contrast to the conventional sliced Wasserstein RBF kernel, we can form an unbiased estimation of USW-RBF kernel via Monte Carlo estimation. Therefore, it is well-suited to stochastic gradient optimization algorithms, and its approximation error decreases at a parametric rate of $\mathcal{O}(L^{-1/2})$ with $L$ Monte Carlo samples. Additionally, we introduce an audio captioning framework based on the unbiased sliced Wasserstein kernel, incorporating stochastic decoding methods to mitigate caption degeneration during the generation process. We conduct extensive quantitative and qualitative experiments on two datasets, AudioCaps and Clotho, to illustrate the capability of generating high-quality audio captions. Experimental results show that our framework is able to increase caption length, lexical diversity, and text-to-audio self-retrieval accuracy.
Understanding and explaining differences between audio recordings is crucial for fields like audio forensics, quality assessment, and audio generation. This involves identifying and describing audio events, acoustic scenes, signal characteristics, and their emotional impact on listeners. This paper stands out as the first work to comprehensively study the task of explaining audio differences and then propose benchmark, baselines for the task. First, we present two new datasets for audio difference explanation derived from the AudioCaps and Clotho audio captioning datasets. Using Large Language Models (LLMs), we generate three levels of difference explanations: (1) concise descriptions of audio events and objects, (2) brief sentences about audio events, acoustic scenes, and signal properties, and (3) comprehensive explanations that include semantics and listener emotions. For the baseline, we use prefix tuning where audio embeddings from two audio files are used to prompt a frozen language model. Our empirical analysis and ablation studies reveal that the naive baseline struggles to distinguish perceptually similar sounds and generate detailed tier 3 explanations. To address these limitations, we propose ADIFF, which introduces a cross-projection module, position captioning, and a three-step training process to enhance the model's ability to produce detailed explanations. We evaluate our model using objective metrics and human evaluation and show our model enhancements lead to significant improvements in performance over naive baseline and SoTA Audio-Language Model (ALM) Qwen Audio. Lastly, we conduct multiple ablation studies to study the effects of cross-projection, language model parameters, position captioning, third stage fine-tuning, and present our findings. Our benchmarks, findings, and strong baseline pave the way for nuanced and human-like explanations of audio differences.
We present MILS: Multimodal Iterative LLM Solver, a surprisingly simple, training-free approach, to imbue multimodal capabilities into your favorite LLM. Leveraging their innate ability to perform multi-step reasoning, MILS prompts the LLM to generate candidate outputs, each of which are scored and fed back iteratively, eventually generating a solution to the task. This enables various applications that typically require training specialized models on task-specific data. In particular, we establish a new state-of-the-art on emergent zero-shot image, video and audio captioning. MILS seamlessly applies to media generation as well, discovering prompt rewrites to improve text-to-image generation, and even edit prompts for style transfer! Finally, being a gradient-free optimization approach, MILS can invert multimodal embeddings into text, enabling applications like cross-modal arithmetic.
Text-to-Audio (TTA) generation is an emerging area within AI-generated content (AIGC), where audio is created from natural language descriptions. Despite growing interest, developing robust TTA models remains challenging due to the scarcity of well-labeled datasets and the prevalence of noisy or inaccurate captions in large-scale, weakly labeled corpora. To address these challenges, we propose CosyAudio, a novel framework that utilizes confidence scores and synthetic captions to enhance the quality of audio generation. CosyAudio consists of two core components: AudioCapTeller and an audio generator. AudioCapTeller generates synthetic captions for audio and provides confidence scores to evaluate their accuracy. The audio generator uses these synthetic captions and confidence scores to enable quality-aware audio generation. Additionally, we introduce a self-evolving training strategy that iteratively optimizes CosyAudio across both well-labeled and weakly-labeled datasets. Initially trained with well-labeled data, AudioCapTeller leverages its assessment capabilities on weakly-labeled datasets for high-quality filtering and reinforcement learning, which further improves its performance. The well-trained AudioCapTeller refines corpora by generating new captions and confidence scores, serving for the audio generator training. Extensive experiments on open-source datasets demonstrate that CosyAudio outperforms existing models in automated audio captioning, generates more faithful audio, and exhibits strong generalization across diverse scenarios.
Automated audio captioning is a task that generates textual descriptions for audio content, and recent studies have explored using visual information to enhance captioning quality. However, current methods often fail to effectively fuse audio and visual data, missing important semantic cues from each modality. To address this, we introduce LAVCap, a large language model (LLM)-based audio-visual captioning framework that effectively integrates visual information with audio to improve audio captioning performance. LAVCap employs an optimal transport-based alignment loss to bridge the modality gap between audio and visual features, enabling more effective semantic extraction. Additionally, we propose an optimal transport attention module that enhances audio-visual fusion using an optimal transport assignment map. Combined with the optimal training strategy, experimental results demonstrate that each component of our framework is effective. LAVCap outperforms existing state-of-the-art methods on the AudioCaps dataset, without relying on large datasets or post-processing. Code is available at https://github.com/NAVER-INTEL-Co-Lab/gaudi-lavcap.
Large Audio-Language Models (LALMs) have demonstrated remarkable performance in tasks involving audio perception and understanding, such as speech recognition and audio captioning. However, their reasoning capabilities - critical for solving complex real-world problems - remain underexplored. In this work, we conduct the first exploration into integrating Chain-of-Thought (CoT) reasoning into LALMs to enhance their reasoning ability across auditory modalities. We evaluate representative CoT methods, analyzing their performance in both information extraction and reasoning tasks across sound, music, and speech domains. Our findings reveal that CoT methods significantly improve performance on easy and medium tasks but encounter challenges with hard tasks, where reasoning chains can confuse the model rather than improve accuracy. Additionally, we identify a positive correlation between reasoning path length and accuracy, demonstrating the potential of scaling inference for advanced instruction-following and reasoning. This study not only highlights the promise of CoT in enhancing LALM reasoning capabilities but also identifies key limitations and provides actionable directions for future research.
Most current captioning systems use language models trained on data from specific settings, such as image-based captioning via Amazon Mechanical Turk, limiting their ability to generalize to other modality distributions and contexts. This limitation hinders performance in tasks like audio or video captioning, where different semantic cues are needed. Addressing this challenge is crucial for creating more adaptable and versatile captioning frameworks applicable across diverse real-world contexts. In this work, we introduce a method to adapt captioning networks to the semantics of alternative settings, such as capturing audibility in audio captioning, where it is crucial to describe sounds and their sources. Our framework consists of two main components: (i) a frozen captioning system incorporating a language model (LM), and (ii) a text classifier that guides the captioning system. The classifier is trained on a dataset automatically generated by GPT-4, using tailored prompts specifically designed to enhance key aspects of the generated captions. Importantly, the framework operates solely during inference, eliminating the need for further training of the underlying captioning model. We evaluate the framework on various models and modalities, with a focus on audio captioning, and report promising results. Notably, when combined with an existing zero-shot audio captioning system, our framework improves its quality and sets state-of-the-art performance in zero-shot audio captioning.
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.
With the emergence of audio-language models, constructing large-scale paired audio-language datasets has become essential yet challenging for model development, primarily due to the time-intensive and labour-heavy demands involved. While large language models (LLMs) have improved the efficiency of synthetic audio caption generation, current approaches struggle to effectively extract and incorporate detailed audio information. In this paper, we propose an automated pipeline that integrates audio-language models for fine-grained content extraction, LLMs for synthetic caption generation, and a contrastive language-audio pretraining (CLAP) model-based refinement process to improve the quality of captions. Specifically, we employ prompt chaining techniques in the content extraction stage to obtain accurate and fine-grained audio information, while we use the refinement process to mitigate potential hallucinations in the generated captions. Leveraging the AudioSet dataset and the proposed approach, we create AudioSetCaps, a dataset comprising 1.9 million audio-caption pairs, the largest audio-caption dataset at the time of writing. The models trained with AudioSetCaps achieve state-of-the-art performance on audio-text retrieval with R@1 scores of 46.3% for text-to-audio and 59.7% for audio-to-text retrieval and automated audio captioning with the CIDEr score of 84.8. As our approach has shown promising results with AudioSetCaps, we create another dataset containing 4.1 million synthetic audio-language pairs based on the Youtube-8M and VGGSound datasets. To facilitate research in audio-language learning, we have made our pipeline, datasets with 6 million audio-language pairs, and pre-trained models publicly available at https://github.com/JishengBai/AudioSetCaps.