论文检索

Naama Rivlin-Angert, Guy Mor-Lan

We present the first large-scale computational study of political delegitimization discourse (PDD), defined as symbolic attacks on the normative validity of political entities. We curate and manually annotate a novel Hebrew-language corpus of 10,410 sentences drawn from parliamentary speeches (1993-2023), Facebook posts, and leading news outlets (2018-2021), of which 1,812 instances (17.4%) exhibit PDD and 642 carry additional annotations for intensity, incivility, target type, and affective framing. We introduce a two-stage classification pipeline, and benchmark finetuned encoder models and decoder LLMs. Our best model (DictaLM 2.0) attains an F_1 of 0.74 for binary PDD detection and a macro-F_1 of 0.67 for classification of delegitimization characteristics. Applying this classifier to longitudinal and cross-platform data, we see a marked rise in PDD over three decades, higher prevalence on social media versus parliamentary debate, greater use by male politicians than by their female counterparts, and stronger tendencies among right-leaning actors, with pronounced spikes during election campaigns and major political events. Our findings demonstrate the feasibility and value of automated PDD analysis for analyzing democratic discourse.

Chutong Meng, Philipp Koehn

We present Speech Vecalign, a parallel speech document alignment method that monotonically aligns speech segment embeddings and does not depend on text transcriptions. Compared to the baseline method Global Mining, a variant of speech mining, Speech Vecalign produces longer speech-to-speech alignments. It also demonstrates greater robustness than Local Mining, another speech mining variant, as it produces less noise. We applied Speech Vecalign to 3,000 hours of unlabeled parallel English-German (En-De) speech documents from VoxPopuli, yielding about 1,000 hours of high-quality alignments. We then trained En-De speech-to-speech translation models on the aligned data. Speech Vecalign improves the En-to-De and De-to-En performance over Global Mining by 0.37 and 0.18 ASR-BLEU, respectively. Moreover, our models match or outperform SpeechMatrix model performance, despite using 8 times fewer raw speech documents.

Haoran Zhao, Robert D. Hawkins

Polite speech poses a fundamental alignment challenge for large language models (LLMs). Humans deploy a rich repertoire of linguistic strategies to balance informational and social goals – from positive approaches that build rapport (compliments, expressions of interest) to negative strategies that minimize imposition (hedging, indirectness).We investigate whether LLMs employ a similarly context-sensitive repertoire by comparing human and LLM responses to English-language scenarios in both constrained and open-ended production tasks.We find that larger models (\ge70B parameters) successfully replicate key effects from the computational pragmatics literature, and human evaluators prefer LLM-generated responses in open-ended contexts. However, further linguistic analyses reveal that models disproportionately rely on negative politeness strategies to create distance even in positive contexts, potentially leading to misinterpretations. While LLMs thus demonstrate an impressive command of politeness strategies, these systematic differences provide important groundwork for making intentional choices about pragmatic behavior in human-AI communication.

Supriti Sinhamahapatra, Jan Niehues

State-of-the-art (SOTA) Automatic Speech Recognition (ASR) systems primarily rely on acoustic information while disregarding additional multi-modal context. However, visual information are essential in disambiguation and adaptation. While most work focus on speaker images to handle noise conditions, this work also focuses on integrating presentation slides for the use cases of scientific presentation.In a first step, we create a benchmark for multi-modal presentation including an automatic analysis of transcribing domain-specific terminology. Next, we explore methods for augmenting speech models with multi-modal information. We mitigate the lack of datasets with accompanying slides by a suitable approach of data augmentation.Finally, we train a model using the augmented dataset, resulting in a relative reduction in word error rate of approximately 34%, across all words and 35%, for domain-specific terms compared to the baseline model.

Binh Nguyen, Shuju Shi, Ryan Ofman, Thai Le

Recent advances in text-to-speech technology have enabled highly realistic voice generation, fueling audio-based deepfake attacks such as fraud and impersonation. While audio anti-spoofing systems are critical for detecting such threats, prior research has predominantly focused on acoustic-level perturbations, leaving **the impact of linguistic variation largely unexplored**. In this paper, we investigate the linguistic sensitivity of both open-source and commercial anti-spoofing detectors by introducing **TAPAS** (Transcript-to-Audio Perturbation Anti-Spoofing), a novel framework for transcript-level adversarial attacks. Our extensive evaluation shows that even minor linguistic perturbations can significantly degrade detection accuracy: attack success rates exceed **60%** on several open-source detector–voice pairs, and the accuracy of one commercial detector drops from **100%** on synthetic audio to just **32%**. Through a comprehensive feature attribution analysis, we find that linguistic complexity and model-level audio embedding similarity are key factors contributing to detector vulnerabilities. To illustrate the real-world risks, we replicate a recent Brad Pitt audio deepfake scam and demonstrate that TAPAS can bypass commercial detectors. These findings underscore the **need to move beyond purely acoustic defenses** and incorporate linguistic variation into the design of robust anti-spoofing systems. Our source code is available at https://github.com/nqbinh17/audio_linguistic_adversarial.

Amirbek Djanibekov, Nurdaulet Mukhituly, Kentaro Inui, Hanan Aldarmaki, Nils Lukas

Speech Language Models (SLMs) enable natural interactions via spoken instructions, which more effectively capture user intent by detecting nuances in speech. The richer speech signal introduces new security risks compared to text-based models, as adversaries can better bypass safety mechanisms by injecting imperceptible noise to speech. We analyze adversarial attacks under white-box access and find that SLMs are substantially more vulnerable to jailbreak attacks, which can achieve a perfect 100% attack success rate in some instances. To improve security, we propose post-hoc patching defenses used to intervene during inference by modifying the SLM’s activations that improve robustness up to 99% with (i) negligible impact on utility and (ii) without any re-training. We conduct ablation studies to maximize the efficacy of our defenses and improve the utility/security trade-off, validated with large-scale benchmarks unique to SLMs.

Tony Woo, Sehun Lee, Kang-wook Kim, Gunhee Kim

Spoken dialogue systems increasingly employ large language models (LLMs) to leverage their advanced reasoning capabilities. However, direct application of LLMs in spoken communication often yield suboptimal results due to mismatches between optimal textual and verbal delivery. While existing approaches adapt LLMs to produce speech-friendly outputs, their impact on reasoning performance remains underexplored. In this work, we propose **Think-Verbalize-Speak**, a framework that decouples reasoning from spoken delivery to preserve the full reasoning capacity of LLMs. Central to our method is *verbalizing*, an intermediate step that translates thoughts into natural, speech-ready text. We also introduce **ReVerT**, a latency-efficient verbalizer based on incremental and asynchronous summarization. Experiments across multiple benchmarks show that our method enhances speech naturalness and conciseness with minimal impact on reasoning. The project page with the dataset and the source code is available at https://yhytoto12.github.io/TVS-ReVerT.

Agostina Calabrese, Tom Sherborne, Björn Ross, Mirella Lapata

Hate speech detection is key to online content moderation, but current models struggle to generalise beyond their training data. This has been linked to dataset biases and the use of sentence-level labels, which fail to teach models the underlying structure of hate speech. In this work, we show that even when models are trained with more fine-grained, span-level annotations (e.g., “artists” is labeled as target and “are parasites” as dehumanising comparison), they struggle to disentangle the meaning of these labels from the surrounding context. As a result, combinations of expressions that deviate from those seen during training remain particularly difficult for models to detect. We investigate whether training on a dataset where expressions occur with equal frequency across all contexts can improve generalisation. To this end, we create U-PLEAD, a dataset of ~364,000 synthetic posts, along with a novel compositional generalisation benchmark of ~8,000 manually validated posts. Training on a combination of U-PLEAD and real data improves compositional generalisation while achieving state-of-the-art performance on the human-sourced PLEAD.

Tianduo Wang, Lu Xu, Wei Lu, Shanbo Cheng

Recent advances in Automatic Speech Recognition (ASR) have been largely fueled by massive speech corpora. However, extending coverage to diverse languages with limited resources remains a formidable challenge. This paper introduces Speech Back-Translation, a a scalable pipeline that improves multilingual ASR models by converting large-scale text corpora into synthetic speech via off-the-shelf text-to-speech (TTS) models. We demonstrate that just tens of hours of real transcribed speech can effectively train TTS models to generate synthetic speech at hundreds of times the original volume while maintaining high quality. To evaluate synthetic speech quality, we develop an intelligibility-based assessment framework and establish clear thresholds for when synthetic data benefits ASR training. Using Speech Back-Translation, we generate more than 500,000 hours of synthetic speech in ten languages and continue pre-training Whisper-large-v3, achieving average transcription error reductions of over 30%. These results highlight the scalability and effectiveness of Speech Back-Translation for enhancing multilingual ASR systems.

Khai Le-Duc, Tuyen Tran, Bach Phan Tat, Nguyen Kim Hai Bui, Quan Dang Anh, Hung-Phong Tran, Thanh Thuy Nguyen, Ly Nguyen, Tuan Minh Phan, Thi Thu Phuong Tran 等

Multilingual speech translation (ST) and machine translation (MT) in the medical domain enhances patient care by enabling efficient communication across language barriers, alleviating specialized workforce shortages, and facilitating improved diagnosis and treatment, particularly during pandemics. In this work, we present the first systematic study on medical ST, to our best knowledge, by releasing MultiMedST, a large-scale ST dataset for the medical domain, spanning all translation directions in five languages: Vietnamese, English, German, French, and Simplified/Traditional Chinese, together with the models. With 290,000 samples, this is the largest medical MT dataset and the largest many-to-many multilingual ST among all domains. Secondly, we present the most comprehensive ST analysis in the field’s history, to our best knowledge, including: empirical baselines, bilingual-multilingual comparative study, end-to-end vs. cascaded comparative study, task-specific vs. multi-task sequence-to-sequence comparative study, code-switch analysis, and quantitative-qualitative error analysis. All code, data, and models are available online: https://github.com/leduckhai/MultiMed-ST.

Tan-Hanh Pham, Le Hoang Nam, Phu-Vinh Nguyen, Chris Ngo, Truong-Son Hy

Visual Language Models have demonstrated remarkable capabilities across various tasks, including visual question answering and image captioning. However, most models rely on text-based instructions, limiting their effectiveness in natural human-machine interactions. Moreover, the quality of language models primarily depends on reasoning and prompting techniques, such as chain-of-thought, which remain underexplored when using speech instructions. To address these challenges, we propose SilVar, an end-to-end multimodal model that leverages speech instructions for reasoning-based visual question answering. Additionally, we investigate reasoning techniques at different levels, including conversational, simple, and complex speech instructions. SilVar is built upon CLIP, Whisper, and LLaMA 3.1-8B, enabling more intuitive interactions by allowing users to provide verbal or text-based instructions. To this end, we introduce a new dataset designed to challenge models with speech-based reasoning tasks for object localization. This dataset enhances the model’s ability to process and explain visual scenes from spoken input, moving beyond simple object recognition to reasoning-based interactions. To our knowledge, SilVar is the first open-source, speech-driven VLM. We believe SilVar will inspire the next generation of multimodal reasoning models, advancing toward expert artificial general intelligence.

AbdelRahim A. Elmadany, Sang Yun Kwon, Hawau Olamide Toyin, Alcides Alcoba Inciarte, Hanan Aldarmaki, Muhammad Abdul-Mageed

Africa’s rich linguistic diversity remains significantly underrepresented in speech technologies, creating barriers to digital inclusion. To alleviate this challenge, we systematically map the continent’s speech space of datasets and technologies, leading to a new comprehensive benchmark SimbaBench for downstream African speech tasks. Using SimbaBench, we introduce the Simba family of models, achieving state-of-the-art performance across multiple African languages and speech tasks. Our benchmark analysis reveals critical patterns in resource availability, while our model evaluation demonstrates how dataset quality, domain diversity, and language family relationships influence performance across languages. Our work highlights the need for expanded speech technology resources that better reflect Africa’s linguistic diversity and provides a solid foundation for future research and development efforts toward more inclusive speech technologies.

Yiheng Jing, Mingming Zhang, Yong Zhuang, Jiacheng Guo, Juan Wang, Xiaoyang Xu, Wenzhe Yi, Keyan Guo, Hongxin Hu

The rapid growth of video platforms has transformed information dissemination and led to an explosion of multimedia content. However, this widespread reach also introduces risks, as some users exploit these platforms to spread hate speech, which is often concealed through complex rhetoric, making hateful video detection a critical challenge. Existing detection methods rely heavily on unimodal analysis or simple feature fusion, struggling to capture cross-modal interactions and reason through implicit hate in sarcasm and metaphor. To address these limitations, we propose HVGuard, the first reasoning-based hateful video detection framework with multimodal large language models (MLLMs). Our approach integrates Chain-of-Thought (CoT) reasoning to enhance multimodal interaction modeling and implicit hate interpretation. Additionally, we design a Mixture-of-Experts (MoE) network for efficient multimodal fusion and final decision-making. The framework is modular and extensible, allowing flexible integration of different MLLMs and encoders. Experimental results demonstrate that HVGuard outperforms all existing advanced detection tools, achieving an improvement of 6.88% to 13.13% in accuracy and 9.21% to 34.37% in M-F1 on two public datasets covering both English and Chinese.

Zhenqi Jia, Rui Liu, Berrak Sisman, Haizhou Li

Conversational Speech Synthesis (CSS) aims to generate speech with natural prosody by understanding the multimodal dialogue history (MDH). The latest work predicts the accurate prosody expression of the target utterance by modeling the utterance-level interaction characteristics of MDH and the target utterance. However, MDH contains fine-grained semantic and prosody knowledge at the word level. Existing methods overlook the fine-grained semantic and prosodic interaction modeling. To address this gap, we propose MFCIG-CSS, a novel Multimodal Fine-grained Context Interaction Graph-based CSS system. Our approach constructs two specialized multimodal fine-grained dialogue interaction graphs: a semantic interaction graph and a prosody interaction graph. These two interaction graphs effectively encode interactions between word-level semantics, prosody, and their influence on subsequent utterances in MDH. The encoded interaction features are then leveraged to enhance synthesized speech with natural conversational prosody. Experiments on the DailyTalk dataset demonstrate that MFCIG-CSS outperforms all baseline models in terms of prosodic expressiveness. Code and speech samples are available at https://github.com/AI-S2-Lab/MFCIG-CSS.

Zhengdong Yang, Zhen Wan, Sheng Li, Chao-Han Huck Yang, Chenhui Chu

Large language models (LLMs) can rewrite the N-best hypotheses from a speech-to-text model, often fixing recognition or translation errors that traditional rescoring cannot. Yet research on generative error correction (GER) has been focusing on monolingual automatic speech recognition (ASR), leaving its multilingual and multitask potential underexplored. We introduce CoVoGER, a benchmark for GER that covers both ASR and speech-to-text translation (ST) across 15 languages and 28 language pairs. CoVoGER is constructed by decoding Common Voice 20.0 and CoVoST-2 with Whisper of three model sizes and SeamlessM4T of two model sizes, providing 5-best lists obtained via a mixture of beam search and temperature sampling. We evaluated various instruction-tuned LLMs, including commercial models in zero-shot mode and open-sourced models with LoRA fine-tuning, and found that the mixture decoding strategy yields the best GER performance in most settings. CoVoGER will be released to promote research on reliable language-universal speech-to-text GER. The code and data for the benchmark are available at https://github.com/N-Orien/CoVoGER.

Bajian Xiang, Shuaijiang Zhao, Tingwei Guo, Wei Zou

End-to-end Large Speech Language Models (LSLMs) have demonstrated impressive conversational generation abilities, yet consistently fall short of traditional pipeline systems on semantic understanding benchmarks. In this work, we reveal through systematic experimentation that although LSLMs lose some text input performance after speech-text alignment training, the performance gap between speech and text inputs is more pronounced, which we refer to as the modality gap. To understand this gap, we analyze both coarse- and fine-grained text and speech representations. At the coarse-grained level, representations of speech and text in deeper layers are found to be increasingly aligned in direction (cosine similarity), while concurrently diverging in magnitude (Euclidean distance). We further find that representation similarity is strongly correlated with the modality gap. At the fine-grained level, a spontaneous token-level alignment pattern between text and speech representations is observed. Based on this, we introduce the Alignment Path Score to quantify token-level alignment quality, which exhibits stronger correlation with the modality gap. Building on these insights, we design targeted interventions on critical tokens through angle projection and length normalization. These strategies demonstrate the potential to improve correctness for speech inputs. Our study provides the first systematic empirical analysis of the modality gap and alignment mechanisms in LSLMs, offering both theoretical and methodological guidance for future optimization.

Zekun Moore Wang, King Zhu, Chunpu Xu, Wangchunshu Zhou, Jiaheng Liu, Yibo Zhang, Jessie Wang, Ning Shi, Siyu Li, Yizhi Li 等

In this paper, we introduce MIO, a novel foundation model built on multimodal tokens, capable of understanding and generating speech, text, images, and videos in an end-to-end, autoregressive manner. While the emergence of large language models (LLMs) and multimodal large language models (MM-LLMs) propels advancements in artificial general intelligence through their versatile capabilities, they still lack true any-to-any understanding and generation. Recently, the release of GPT-4o has showcased the remarkable potential of any-to-any LLMs for complex real-world tasks, enabling omnidirectional input and output across images, speech, and text. However, it is closed-source and does not support the generation of multimodal interleaved sequences. To address this gap, we present MIO, which is trained on a mixture of discrete tokens across four modalities using causal multimodal modeling. MIO undergoes a four-stage training process: (1) alignment pre-training, (2) interleaved pre-training, (3) speech-enhanced pre-training, and (4) comprehensive supervised fine-tuning on diverse textual, visual, and speech tasks. Our experimental results indicate that MIO exhibits competitive, and in some cases superior, performance compared to previous dual-modal baselines, any-to-any model baselines, and even modality-specific baselines. Moreover, MIO demonstrates advanced capabilities inherent to its any-to-any feature, such as interleaved video-text generation, chain-of-visual-thought reasoning, visual guideline generation, instructional image editing, etc.

Nathan Roll, Calbert Graham, Yuka Tatsumi, Kim Tien Nguyen, Meghan Sumner, Dan Jurafsky

Human listeners readily adjust to unfamiliar speakers and language varieties through exposure, but do these adaptation benefits extend to state-of-the-art spoken language models (SLMs)? We introduce a scalable framework that allows for in-context learning (ICL) in Phi-4 Multimodal (Phi-4-MM) using interleaved task prompts and audio-text pairs, and find that as few as 12 example utterances (~50 seconds) at inference time reduce word error rates by a relative 19.7% (1.2 pp.) on average across diverse English corpora. These improvements are most pronounced in low-resource varieties, when the context and target speaker match, and when more examples are provided—though scaling our procedure yields diminishing marginal returns to context length. Overall, we find that our novel ICL adaptation scheme (1) reveals a similar performance profile to human listeners, and (2) demonstrates consistent improvements to automatic speech recognition (ASR) robustness across diverse speakers and language backgrounds. While adaptation succeeds broadly, significant gaps remain for certain varieties, revealing where current models still fall short of human flexibility. We release our prompts and code on GitHub.

Jianxing Yu, Gou Zihao, Chen Li, Zhisheng Wang, Peiji Yang, Wenqing Chen, Jian Yin

This paper focuses on generating speech with the acoustic style that meets users’ needs based on their open-domain instructions. To control the style, early work mostly relies on pre-defined rules or templates. The control types and formats are fixed in a closed domain, making it hard to meet diverse needs of users. One solution is to resort to instructions in free text to guide the generation. Current work mainly studies the instructions that clearly specify the acoustic styles, such as low pitch and fast speed. However, the instructions are complex, some even vague and abstract, such as “Generate a voice of a woman who is heartbroken due to a breakup. It is hard to infer this implicit style by traditional matching-based methods. To address this problem, we propose a new controllable model. It first utilizes multimodal LLMs with knowledge-augmented techniques to infer the desired speech style from the instructions. The powerful language understanding ability of LLMs can help us better elicit the implicit style factors from the instruction. By using these factors as a control condition, we design a diffusion-based generator adept at finely adjusting speech details. That offers higher flexibility to meet complex users’ needs. Next, we verify the output speech from three aspects, i.e., consistency of decoding state, mel-spectrogram, and instruction style. This verified feedback can inversely optimize the generator. Extensive experiments are conducted on three popular datasets. The results show the effectiveness and good controllability of our approach.

Anuj Diwan, Zhisheng Zheng, David Harwath, Eunsol Choi

We introduce Paralinguistic Speech Captions (ParaSpeechCaps), a large-scale dataset that annotates speech utterances with rich style captions. While rich abstract tags (e.g. guttural, nasal, pained) have been explored in small-scale human-annotated datasets, existing large-scale datasets only cover basic tags (e.g. low-pitched, slow, loud). We combine off-the-shelf text and speech embedders, classifiers and an audio language model to automatically scale rich tag annotations for the first time. ParaSpeechCaps covers a total of 59 style tags, including both speaker-level intrinsic tags and utterance-level situational tags. It consists of 282 hours of human-labelled data (PSC-Base) and 2427 hours of automatically annotated data (PSC-Scaled). We finetune Parler-TTS, an open-source style-prompted TTS model, on ParaSpeechCaps, and achieve improved style consistency (+7.9% Consistency MOS) and speech quality (+15.5% Naturalness MOS) over the best performing baseline that combines existing rich style tag datasets. We ablate several of our dataset design choices to lay the foundation for future work in this space. Our dataset, models and code are released at https://github.com/ajd12342/paraspeechcaps .