论文检索

Applications · Computer Vision

Foivos Paraperas Papantoniou, Stathis Galanakis, Rolandos Alexandros Potamias, Bernhard Kainz, Stefanos Zafeiriou

This paper presents STARCaster, an identity-aware spatio-temporal video diffusion model that addresses both speech-driven portrait animation and free-viewpoint talking portrait synthesis, given an identity embedding or reference image, within a unified framework. Existing 2D speech-to-video diffusion models depend heavily on reference guidance, leading to limited motion diversity. At the same time, 3D-aware animation typically relies on inversion through pre-trained tri-plane generators, which often leads to imperfect reconstructions and identity drift. We rethink reference- and geometry-based paradigms in two ways. First, we deviate from strict reference conditioning at pre-training by introducing softer identity constraints. Second, we address 3D awareness implicitly within the 2D video domain by leveraging the inherent multi-view nature of video data. STARCaster adopts a compositional approach progressing from ID-aware motion modeling, to audio-visual synchronization via lip reading-based supervision, and finally to novel view animation through temporal-to-spatial adaptation. To overcome the scarcity of 4D audio-visual data, we propose a decoupled learning approach in which view consistency and temporal coherence are trained independently. Comprehensive evaluations demonstrate that STARCaster generalizes effectively across tasks and identities, consistently surpassing prior approaches in different benchmarks.

Applications · Language, Speech and Dialog

Yiqun Sun, Qiang Huang, Anthony Tung, Jun Yu

**This position paper argues that text embedding research should move beyond surface meaning and embrace implicit semantics as a central modeling objective.** Text embeddings are a foundational component of modern NLP, underpinning a wide range of applications and driving sustained research progress. Despite rapid progress, most embedding models remain narrowly focused on surface-level semantics, whereas linguistic theory emphasizes that much of human meaning is implicit, shaped by pragmatics, speaker intent, and sociocultural context. Current embedding models are typically trained on datasets that lack such depth and evaluated using benchmarks that reward surface similarity. As a result, they struggle with tasks that require interpretive reasoning, stance recognition, or socially grounded understanding. Our pilot study makes this limitation explicit, showing that even state-of-the-art embeddings achieve only marginal improvements over simple lexical baselines on tasks probing implicit semantics. We therefore call for a paradigm shift: embedding research should prioritize linguistically grounded and diverse training data, develop benchmarks that probe deeper semantic understanding, and treat implicit meaning as a core modeling objective to better align embeddings with real-world language complexity.

Applications · Language, Speech and Dialog

Qiuchen Wang, Shihang Wang, Yu Zeng, Qiang Zhang, Fanrui Zhang, Zhuoning Guo, Bosi Zhang, Wenxuan Huang, Lin Chen, Zehui Chen 等

Effectively retrieving, reasoning, and understanding multimodal information remains a critical challenge for agentic systems. Traditional Retrieval-augmented Generation (RAG) methods rely on linear interaction histories, which struggle to handle long-context tasks, especially those involving information-sparse yet token-heavy visual data in iterative reasoning scenarios. To bridge this gap, we introduce VimRAG, a framework tailored for multimodal Retrieval-augmented Reasoning across text, images, and videos. Inspired by our systematic study, we model the reasoning process as a dynamic directed acyclic graph that structures the agent states and retrieved multimodal evidence. Building upon this structured memory, we introduce a Graph-Modulated Visual Memory Encoding mechanism, with which the significance of memory nodes is evaluated via their topological position, allowing the model to dynamically allocate high-resolution tokens to pivotal evidence while compressing or discarding trivial clues. To implement this paradigm, we propose a Graph-Guided Policy Optimization strategy. This strategy disentangles step-wise validity from trajectory-level rewards by pruning memory nodes associated with redundant actions, thereby facilitating fine-grained credit assignment. Extensive experiments demonstrate that VimRAG consistently achieves state-of-the-art performance on diverse multimodal RAG benchmarks.

Applications · Language, Speech and Dialog

Doyoung Kim, Youngjun Lee, Joeun Kim, Jihwan Bang, Hwanjun Song, Susik Yoon, Jae-Gil Lee

Conversational query reformulation (CQR) has become indispensable for improving retrieval in dialogue-based applications. However, existing approaches typically rely on reference passages for optimization, which are **impractical** to acquire in real-world scenarios. To address this limitation, we introduce a novel **reference-free** preference optimization framework ***DualReform*** that generates **pseudo reference passages** from **commonly-encountered** conversational datasets containing only queries and responses. DualReform attains this goal through two key innovations: (1) **response-based inference**, where responses serve as proxies to infer pseudo reference passages, and (2) **response refinement via the dual-role of CQR**, where a CQR model refines responses based on the shared objectives between response refinement and CQR. Despite not relying on reference passages, ***DualReform*** achieves 96.9--99.1% of the retrieval accuracy attainable only with reference passages and surpasses the state-of-the-art method by up to 31.6%.

Applications · Language, Speech and Dialog

Minchan Kwon, Sung Hyun Baek, Minseo Kim, Jaemyung Yu, Dongyoon Han, Junmo Kim

Large Language Model Red-Teaming, which proactively identifies vulnerabilities of large language models, is an essential process for ensuring safety. Finding effective and diverse attacks in red team activities is important, but achieving both is challenging. Generative Flow Networks (GFN) that perform distribution matching are a promising method, but they are notorious for training instability and mode collapse. In particular, unstable reward functions in red team activities accelerate mode collapse. We propose Stable-GFN (S-GFN), which eliminates Z estimation in GFN and reduces training instability. S-GFN avoids Z-estimation through pairwise comparisons and employs a robust masking methodology against noisy rewards. Additionally, we propose a fluency stabilizer to prevent the model from getting stuck in local optima that produce gibberish. S-GFN provides more stable training while maintaining the optimal policy of GFN. We demonstrate the overwhelming attack performance and diversity of S-GFN across various settings.

Applications · Language, Speech and Dialog

Haohao Luo, Zexi Li, Yuexiang Xie, Wenhao Zhang, Yaliang Li, Ying Shen

Deep Research (DR) agents extend Large Language Models (LLMs) beyond parametric knowledge by autonomously retrieving and synthesizing evidence from large web corpora into long-form reports, enabling a long-horizon agentic paradigm. However, unlike real-time conversational assistants, DR is computationally expensive and time-consuming, creating an autonomy-interaction dilemma: high autonomy on ambiguous user queries often leads to prolonged execution with unsatisfactory outcomes. To address this, we propose IntentRL, a framework that trains proactive agents to clarify latent user intents before starting long-horizon research. To overcome the scarcity of open-ended research data, we introduce a scalable pipeline that expands a few seed samples into high-quality dialogue turns via a shallow-to-deep intent refinement graph. We further adopt a two-stage reinforcement learning (RL) strategy: Stage I applies RL on offline dialogues to efficiently learn general user-interaction behavior, while Stage II uses the trained agent and a user simulator for online rollouts to strengthen adaptation to diverse user feedback. Extensive experiments show that IntentRL significantly improves both intent hit rate and downstream task performance, outperforming the built-in clarify modules of closed-source DR agents and proactive LLM baselines.

Applications · Language, Speech and Dialog

Sang-Hoon Lee, Ha-Yeong Choi

Representation alignment (REPA) has been investigated to accelerate diffusion training, but we observe that regularizing intermediate representations in diffusion Transformers (DiT) may implicitly entangle latents and limit generative capacity. To address this issue, we propose ReGen, a hierarchical multi-prompt representation generation framework that jointly estimates multiple vector fields for both representations and data within a single diffusion model. We further introduce generalized flow matching (GFM) to improve the generalization of conditional flow matching (CFM). We validate ReGen on single-stage waveform diffusion models including neural audio codec and Wave-VAE. ReGen significantly improves waveform generation quality from highly compressed latent representations at 12.5 Hz. We also present ReGenVoice, a latent diffusion model (LDM)-based text-to-speech model that achieves strong speech intelligibility (WER) and speaker similarity (SIM) with a small dataset. Moreover, operating the LDM at 6.25 Hz with rich semantic and acoustic latent representation enables efficient training and sampling, requiring only 1 day of training on 4 GPUs and fast inference with an RTF of 0.08.

Deep Learning · Foundation Models

Yi Zhu, Brahmi Dwivedi, Jayaram Raghuram, Surya Koppisetti

Existing voice deepfake detection and localization models rely heavily on representations extracted from speech foundation models (SFMs). However, downstream finetuning has now reached a state of diminishing returns. In this paper, we shift the focus to pretraining and propose a novel recipe that combines *bottleneck masked embedding prediction with flow-matching based spectrogram reconstruction*. The outcome, *Alethia*, is the first foundational audio encoder for various voice deepfake detection and localization tasks. We evaluate on 5 different tasks with 56 benchmark datasets, and note *Alethia* significantly outperforms state-of-the-art SFMs with superior robustness to real-world perturbations and zero-shot generalization to unseen domains (e.g., singing deepfakes). We also demonstrate the limitation of discrete targets in masked token prediction, and show the importance of *continuous embedding* prediction and *generative pretraining* for capturing deepfake artifacts.

Applications · Language, Speech and Dialog

Linjuan Wu, Ruiqi Zhang, Xinze Lyu, Ye Guo, Daoxin Zhang, Zhe Xu, Yao Hu, Yixin Cao, Yongliang Shen, Weiming Lu

Social media platforms enable large-scale cross-lingual communication, yet translating user-generated content (UGC) remains challenging due to its informal style, culture-laden expressions, and interaction-driven nuances. While recent LLMs have advanced translation quality, existing benchmarks and metrics often overlook whether translations preserve intended meaning and cultural resonance in real-world contexts. In this work, we introduce **CULTURE-MT**, a benchmark for social media translation that explicitly emphasizes **CUL**tural **T**ransmission and **U**GC-specific emotion **RE**sonance. CULTURE-MT comprises 1,002 Chinese-to-English UGC notes spanning 14 domains, systematically categorized into four types based on culture-loaded symbols and linguistic styles. We also construct UGC-oriented training data to fine-tune Qwen3-8B and Qwen3-32B as strong baselines. We propose the **cultural effectiveness** criterion and train a related JUDGER model that jointly assesses expression accuracy and cultural adaptability. Evaluating 15 models, we find that standard automatic metrics are largely insensitive to cultural effectiveness. Our work establishes a comprehensive framework for evaluating and advancing UGC translation, and will provide an open evaluation platform to support future research in culturally effective UGC translation.

Applications · Language, Speech and Dialog

Mengyi DENG, Zhiwei Li, Xin Li, Tingyu ZHU, Ying Zhao, Zhijiang Guo, Wei Wang

Large Language Model (LLM) agents often operate under underspecified user instructions, where latent uncertainty over user intent leads to erroneous tool actions. To address this challenge, we propose a goal-oriented clarification framework that aligns clarification behavior with ambiguity resolution. Central to our approach is the Information Gain Reward, a metric that quantifies the utility of clarification questions by measuring the Bayesian belief update towards the ground-truth goal induced by the clarification exchange. We train the clarifier (LLM) using this reward to optimize for high information gain, ensuring that clarifications effectively reduce uncertainty and improve task completion within the agent-tool-user environment. We validate our framework within a clarification-enhanced $\tau$-Bench environment, conducting cross-agent evaluations across five heterogeneous backbones. Empirical results demonstrate that our method consistently improves the success rate by 3.7\% over the no-clarification baseline, incurring small overhead via surgical interventions (averaging only 0.3 additional steps). Our code and data are available at the anonymous repository: \url{https://anonymous.4open.science/r/Information-Gain-Clarifier-AD68/}.

Social Aspects · Safety

Yudong Yang, Xuezhen Zhang, Zhifeng Han, Siyin Wang, Jimin Zhuang, Zengrui Jin, Jing Shao, Guangzhi Sun, Chao Zhang

Recent progress in large language models (LLMs) has enabled understanding of both speech and non-speech audio, but has also exposed new safety risks arising from complex audio inputs that are inadequately handled by current safeguards. We introduce SACRED-Bench (Speech–Audio Composition for RED-teaming) to evaluate the robustness of LLMs under complex audio-based attacks. Unlike existing perturbation-based methods that rely on noise optimization or white-box access, SACRED-Bench exploits speech–audio composition to enable effective black-box attacks. SACRED-Bench adopts three composition mechanisms: (a) speech overlap, (b) multi-speaker dialogue, and (c) mixtures of speech and non-speech audio. These mechanisms focus on evaluating safety in settings where benign and harmful intents co-occur within a single auditory scene. Moreover, questions in SACRED-Bench are designed to implicitly refer to content in the audio, such that no explicit harmful information appears in the text prompt alone. Experiments demonstrate that even Gemini 2.5 Pro, a state-of-the-art proprietary LLM with safety guardrails fully enabled, still exhibits a 66% attack success rate. To bridge this gap, we propose SALMONN-Guard, the first guard model that jointly inspects speech, audio, and text for safety judgments, reducing the attack success rate to 20\%. Our results highlight the need for audio-aware defenses to ensure the safety of multimodal LLMs.

Applications · Language, Speech and Dialog

Mugeng Liu, Shuoqi Li, Yixuan Zhang, Yun Ma

In the agentic web era, LLM-based agents increasingly invoke web services as tools, yet most interfaces are still exposed as \emph{static endpoints}. As tasks grow into long-horizon workflows with loops, conditionals, joins, and retries, agents must externalize control flow into stepwise calls and multi-round reasoning, causing excessive network turns, over-/under-fetching, and brittle recovery under partial failures. We argue that the bottleneck is representational: endpoint sequences are a poor interface for expressing tool intent. We present ToolPro, which represents an agent’s tool intent as an \emph{executable tool program}---a composable program that compactly encodes multi-step service interactions with explicit effect types. ToolPro combines (i) constraint-guided program construction with compiler/runtime feedback to improve executability of LLM-produced code, (ii) effect-aware replay that provides exactly-once semantics for state-modifying calls across iterative repair and re-execution, and (iii) a profile-driven consolidation policy that decides when program execution outperforms stepwise calling. We instantiate ToolPro over MCP-style services with WebAssembly sandboxing. Across three real applications and six workflows, ToolPro reduces end-to-end latency by up to 37.2\% and client-side traffic by up to 85.3\%, with larger gains under higher network latency and increased workflow complexity. We believe that ToolPro lays a foundation for agent-facing service interfaces in the emerging agentic web.

Applications · Language, Speech and Dialog

Masataka Yoneda, Yusuke Matsushita, Go Kamoda, Kohei Suenaga, Takuya Akiba, Masaki Waga, Sho Yokoi

We present an ultra-fast and flexible search algorithm that enables search over trillion-scale natural language corpora in under 0.3 seconds while handling semantic variations (substitution, insertion, and deletion). Our approach employs string matching based on suffix arrays that scales well with corpus size. To mitigate the combinatorial explosion induced by the semantic relaxation of queries, our method is built on two key algorithmic ideas: fast exact lookup enabled by a disk-aware design, and dynamic corpus-aware pruning. We theoretically show that the proposed method suppresses exponential growth in the search space with respect to query length by leveraging statistical properties of natural language. In experiments on FineWeb-Edu (1.4T tokens), we show that our method achieves significantly lower search latency than existing methods: infini-gram, infini-gram mini, and SoftMatcha. As a practical application, we demonstrate that our method identifies benchmark contamination in training corpora, unidentified by existing approaches. We also provide an online demo of fast, soft search across corpora in six languages.

Applications · Language, Speech and Dialog

Tom Labiausse, Romain Fabre, Yannick Estève, Alexandre Défossez, Neil Zeghidour

Simultaneous speech translation is the task of translating source speech into a target language in real-time. Given that the dependencies between source and target words are non-monotonic (e.g. the word order can change between German and English), this means learning to jointly align and translate. This task has been traditionally tackled through supervised training on aligned data, and as collecting such data is challenging, this relies on synthetic data with automatic alignment. The latter relies on heuristics that are language-specific and suboptimal. We instead propose Hibiki-Zero, a model for simultaneous speech translation trained without word-level alignments between source and target speech. To do so, we train on sentence-level aligned data so that the model learns to perform speech translation but with high latency. We then introduce a novel reinforcement learning strategy relying on GRPO to optimize the translation latency of the model while retaining its translation capabilities. After supervised and post-training, Hibiki-Zero performs multilingual simultaneous translation with state-of-the-art translation accuracy, latency, voice transfer and naturalness across five X-to-English tasks. Moreover, we demonstrate that our model can be easily finetuned to support another language as input with less than 1000h of speech data. We provide examples ([hibiki-zero-s2st.github.io](https://hibiki-zero-s2st.github.io)) as well as models and release a benchmark containing 15h of multilingual data for speech translation evaluation.

Tom Labiausse, Romain Fabre, Yannick Estève, Alexandre Défossez, Neil Zeghidour

Simultaneous speech translation is the task of translating source speech into a target language in real-time. Given that the dependencies between source and target words are non-monotonic (e.g. the word order can change between German and English), this means learning to jointly align and translate. This task has been traditionally tackled through supervised training on aligned data, and as collecting such data is challenging, this relies on synthetic data with automatic alignment. The latter relies on heuristics that are language-specific and suboptimal. We instead propose Hibiki-Zero, a model for simultaneous speech translation trained without word-level alignments between source and target speech. To do so, we train on sentence-level aligned data so that the model learns to perform speech translation but with high latency. We then introduce a novel reinforcement learning strategy relying on GRPO to optimize the translation latency of the model while retaining its translation capabilities. After supervised and post-training, Hibiki-Zero performs multilingual simultaneous translation with state-of-the-art translation accuracy, latency, voice transfer and naturalness across five X-to-English tasks. Moreover, we demonstrate that our model can be easily finetuned to support another language as input with less than 1000h of speech data. We provide examples ([hibiki-zero-s2st.github.io](https://hibiki-zero-s2st.github.io)) as well as models and release a benchmark containing 15h of multilingual data for speech translation evaluation.

Yizhong Geng, Yanliang Li, Jinghan Yang, Tianhan Jiang, Boxun An, Ya Li, Xiaoyu Shen

Spoken Language Models (SLMs) revolutionize speech synthesis by bypassing traditional linguistic front-ends, yet they remain limited by the digital resource disparities across languages. We investigate these challenges within the Southeast Asian linguistic landscape, using the phonetically complex Thai and data-scarce Lao as representative cases for low-resource SLMs. Scaling experiments reveal that reliance on synthetic data triggers a Stability-Expressivity Gap, characterized by a non-monotonic degradation we term Synthetic Erosion. To bridge this gap, we propose two self-alignment frameworks. Disentanglement-Guided Self-Alignment (DGSA) recovers expressivity for complex languages by exploiting prosody-timbre separation. For regimes where authentic references are exceptionally limited, Temperature-Driven Self-Critique (TDSC) stabilizes generation through automated exploration and filtering. Our methods achieve state-of-the-art results, including the first zero-shot voice cloning capability for Lao, establishing a scalable pathway for high-fidelity synthesis across the global linguistic long-tail. Audio Samples are available at: \url{https://anonymous.4open.science/api/repo/multilantts-demo-EEF6/file/index.html?v=2de23271}.

Applications · Language, Speech and Dialog

Yong Ren, Jingbei Li, Haiyang Sun, Yujie Chen, Cheng Yi, Yechang Huang, Hao Gu, Ye Bai, Xuerui Yang

Recent advances in Large Audio Language Models (LALMs) have extended Text-to-Speech (TTS) to interactive role-play scenarios, which demand high expressiveness and strict adherence to role-play instructions. However, existing models struggle to maintain stylistic consistency with character profiles and scene descriptions across multi-turn dialogues. A critical bottleneck is the lack of objective metrics for quantifying speaking style. To bridge this gap, we propose **Mean Continuation Log-Probability (MCLP)** as both an evaluation metric and a reward signal, validated on LALM-based Role-Play TTS (RP-TTS) tasks. Critically, we leverage the In-Context Learning capability of pre-trained LALMs to formulate MCLP via a continuation log-probability prediction. This metric quantifies stylistic consistency by measuring the likelihood of the ground-truth speech conditioned on the generated speech. Furthermore, we employ MCLP as a reinforcement learning reward to enhance the style alignment between generated speech and Role-Play instructions. To facilitate evaluation, we construct an RP-TTS dataset with rich scene and character annotations. Experimental results demonstrate that our method significantly outperforms strong LALM baselines on both objective and subjective metrics. Audio samples are available at https://mclp-tts.github.io/mclp/.

Applications · Language, Speech and Dialog

Jing Ye, Yiwen Duan, Yonghong Yu, Victor Ma, Yang Gao, Xing Chen

SQL is central to enterprise data engineering, yet generating fully correct SQL code in a single attempt remains difficult—even for experienced developers and advanced \ttsql LLMs—often requiring multiple debugging iterations. We introduce \textbf{\ourbench}, the first benchmark for enterprise-level SQL reasoning and debugging. Our benchmark is built upon two key innovations: (1) an \textbf{automated construction workflow} that employs reverse engineering to systematically inject realistic bugs into large-scale SQL code, enabling scalable and diverse benchmark generation; and (2) an \textbf{execution-free evaluation framework} tailored for enterprise settings, providing fast, accurate, and resource-efficient assessment. \ourbench comprises $469$ \ourbenchsyn queries featuring syntax errors with explicit error messages, and $516$ \ourbenchsem queries targeting semantic errors where codes fails to meet user intent. The queries are highly complex, averaging over $140$ lines, and featuring deep and wide abstract syntax trees (average width $>11$, depth $>8.7$). Evaluation of nearly $30$ LLMs reveals a substantial performance gap: the best-performing model, Claude-4-Sonnet, achieves only $36.46\%$ accuracy on \ourbenchsyn and $32.17\%$ on \ourbenchsem, while most models score below $20\%$. We further explore four solution strategies, identify key challenges, and outline promising directions for enterprise SQL debugging with LLMs.

Applications · Language, Speech and Dialog

Xiang Li, Yixuan Zhou, Xie, Zhiyong Wu, Hui Wang

Neural speech codecs based on Vector-Quantized VAEs (VQ-VAEs) are core audio tokenizers for speech LLMs, yet their reconstruction fidelity is bottlenecked by quantization error. Instead of modifying the quantizer or increasing model capacity—common approaches that complicate downstream language modeling—we introduce self-guidance, a simple yet general training principle that enhances the decoder's robustness to quantization error. Our core idea is to align the decoder's internal feature manifolds when processing both the quantized tokens and their original continuous embeddings, using a lightweight feature-mapping loss. This requires minimal training overhead and no inference-time changes. Applied to XCodec2, self-guidance improves all reconstruction metrics, achieving state-of-the-art low-bitrate performance. It generalizes across codebook sizes, quantizer types, and network architectures, demonstrating value as a universal codec enhancer. Notably, it enables a 4× codebook reduction without fidelity loss, which downstream TTS experiments show significantly improves LLM-based synthesis by simplifying the token modeling space. Self-guidance thus establishes an efficient, broadly applicable method for high-fidelity neural audio coding.

General Machine Learning · Evaluation

Maris Basha, Anja T. Zai, Sabine Stoll, Richard Hahnloser

General-purpose audio representations aim to map acoustically variable instances of the same event to nearby points, resolving content identity in a zero-shot setting. Unlike supervised classification benchmarks that measure adaptability via parameter updates, we introduce VocSim, a training-free benchmark probing the intrinsic geometric alignment of frozen embeddings. VocSim aggregates 125k single-source clips from 19 corpora spanning human speech, animal vocalizations, and environmental sounds. By restricting to single-source audio, we isolate content representation from the confound of source separation. We evaluate embeddings using Precision@k for local purity and the Global Separation Rate (GSR) for point-wise class separation. To calibrate GSR, we report lift over an empirical permutation baseline. Across diverse foundation models, a simple pipeline, frozen Whisper encoder features, time–frequency pooling, and label-free PCA, yields strong zero-shot performance. However, VocSim also uncovers a consistent generalization gap. On blind, low-resource speech, local retrieval drops sharply. While performance remains statistically distinguishable from chance, the absolute geometric structure collapses, indicating a failure to generalize to unseen phonotactics. As external validation, our top embeddings predict avian perceptual similarity, improve bioacoustic classification, and achieve state-of-the-art results on the HEAR benchmark. We posit that the intrinsic geometric quality measured here proxies utility in unlisted downstream applications. We release data, code, and a public leaderboard to standardize the evaluation of intrinsic audio geometry.