多模态大模型

Acoustic environments often contain multiple overlapping sound events, and the same acoustic scene can be described using diverse textual expressions, making audio-text alignment inherently ambiguous. This paper proposes a probabilistic audio-language pretraining framework to model many-to-many correspondence ambiguity in audio-text alignment. Unlike conventional contrastive methods that learn deterministic point embeddings, our approach represents each modality as a distribution and learns uncertainty-aware cross-modal alignment. Rather than relying on masking-based uncertainty simulation, we mix audio-text pairs to create overlapping sounds that better reflect real acoustic mixtures and capture semantic inclusion relations among sound events. We further introduce a multi-level inclusion loss to enforce representations consistent with these relations. Experiments on audio-text retrieval benchmarks show that the proposed method outperforms deterministic baselines.

Self-supervised speech encoders are often fine-tuned with language supervision, which can overlook geographical variation. To understand the learned representations under joint supervision of language and district compared to language-only supervision, we fine-tune Whisper-base and Wav2Vec2.0-base for classification tasks with joint language-district (386 classes) and language-only classification (60 languages). The language-district supervision improves district discrimination conditioned on language in the embedding space while strong marginal language classification. We analyze the structure of the learned embeddings using Normalized Conditional Mutual Information (NCMI), showing that language-district supervision produces global language clusters with structured within language subclusters aligned to district variation, enhancing geographical separability without degrading language-level organization.

Self-supervised speech encoders are predominantly trained by predicting discrete hard cluster IDs at masked positions, a recipe that collapses acoustic ambiguity at category boundaries and requires interrupting training to re-cluster the entire corpus between iterations. We introduce S-JEPA, a JEPA-style encoder-predictor pair trained to match the soft posteriors of a Gaussian Mixture Model at masked positions via KL divergence. Training runs as one continuous optimization trajectory in two phases: a fixed GMM over MFCC features, then an online GMM over encoder features, with the input layer selected adaptively from a label-free signal, removing both the offline re-cluster step and the hand-tuned choice of which transformer layer to cluster on. Under the SUPERB protocol, S-JEPA achieves the lowest WER among evaluated SSL methods below 90M parameters and matches HuBERT-Base on emotion recognition at roughly half its parameter count, establishing a new Pareto frontier without offline re-clustering or teacher distillation. An analysis of the predictor's per-frame entropy on held-out speech reveals a bimodal distribution with a substantial minority of frames near the entropy of a perfect two-cluster tie, providing direct empirical evidence that the soft-target objective preserves the acoustic ambiguity that hard targets would collapse. Code is available at https://github.com/gioannides/s-jepa.

Code-switch (CS) Automatic Speech Recognition (ASR) remains challenging due to limited availability of high quality CS text-speech pairs for training. Although synthetic data augmentation via Text-to-speech (TTS) has been explored, existing CS TTS approaches primarily optimise reconstruction fidelity and do not explicitly enforce language-boundary consistency, thereby limiting their effectiveness for CS ASR augmentation. This paper proposes a code-mixing guided preference-learning framework that steers synthetic speech generation toward improved code-switching fidelity using the Code Mixing Index (CMI). Experiments on the SEAME Mandarin-English conversational corpus demonstrate that the proposed method enhances the utility of synthetic data for ASR fine-tuning. Specifically, when fine-tuning Whisper Large, the proposed approach reduces Mixed Error Rate (MER) from 12.1%/17.8% to 8.9%/14.2% on the DevMAN and DevSGE sets, respectively.

Recent flow-matching text-to-speech (TTS) models, such as F5-TTS, rely on a reference transcript at inference time, obtained from an external ASR system. This dependency makes zero-shot TTS brittle for accented or dysarthric speakers, precisely the scenarios where it is most needed. Moreover, we find that text-based reference conditioning can propagate atypical acoustic patterns from atypical speech into synthesis, even when ground-truth transcripts are available. To address this, we propose RTFree-F5, which replaces the reference transcript with continuous self-supervised speech representations mapped into F5-TTS's text-conditioning space via a lightweight adapter, while reusing the pretrained checkpoint. On dysarthric speech, RTFree-F5 reduces WER from 24.6% to 10.4%, surpassing even the ground-truth reference transcript baselines, while improving naturalness and remaining competitive on standard benchmarks without requiring any reference transcript.

Classifier guidance is a way to control diffusion generation by using a noise-conditioned classifier to steer the sampling process toward a target class. One drawback of classifier guidance is that it requires two separately trained models: a classifier and a diffusion model. We therefore study a more compact alternative in which a conventionally trained speech classifier is repurposed as the backbone for diffusion generation. Starting from a frozen noise-conditioned classifier in log-Mel space, we attach a lightweight subnetwork that reuses intermediate classifier representations and train only this subnetwork under a Denoising Score Matching objective. Our work shows that a pretrained classifier can be repurposed for conditional generation, providing an appealing bridge between discriminative modeling and conditional speech synthesis resulting in high speech quality within a single-backbone model, with reduced memory footprint and computational cost.

Satellite communications face severe bottlenecks in supporting high-fidelity synchronized audiovisual services, as conventional schemes struggle with cross-modal coherence under fluctuating channel conditions, limited bandwidth, and long propagation delays. To address these limitations, this paper proposes an adaptive multimodal semantic transmission system tailored for satellite scenarios, aiming for high-quality synchronized audiovisual reconstruction under bandwidth constraints. Unlike static schemes with fixed modal priorities, our framework features a dual-stream generative architecture that flexibly switches between video-driven audio generation and audio-driven video generation. This allows the system to dynamically decouple semantics, transmitting only the most important modality while employing cross-modal generation to recover the other. To balance reconstruction quality and transmission overhead, a dynamic keyframe update mechanism adaptively maintains the shared knowledge base according to wireless scenarios and user requirements. Furthermore, a large language model based decision module is introduced to enhance system adaptability. By integrating satellite-specific knowledge, this module jointly considers task requirements and channel factors such as weather-induced fading to proactively adjust transmission paths and generation workflows. Simulation results demonstrate that the proposed system significantly reduces bandwidth consumption while achieving high-fidelity audiovisual synchronization, improving transmission efficiency and robustness in challenging satellite scenarios.

Stuttering detection and classification using deep learning methods has the potential to improve the process of stuttering severity assessment. Most stuttering classification datasets provide clip-level labels, making them unsuitable for fine-grained frame-level classification needed to determine the duration of individual stuttering dysfluencies. To overcome this challenge, we present a multiple instance neural network architecture based on fine-tuned wav2vec 2.0, WavLM and Whisper encoders. We apply instance- and embedding-based multiple instance learning approaches to train models on a clip-level dataset for both clip-level and frame-level stuttering classification tasks. Our results show a 23% improvement in frame-level F1 score and between 2% and 9% in clip-level F1 score, demonstrating the ability of our models to utilize clip-level data for frame-level segmentation.

Most generative speech enhancement methods rely on explicit time-step embeddings for temporal conditioning. In this paper, we propose the Autonomous Rectified Flow framework, which challenges the necessity of such conditioning. Using a linear interpolation path, we show that the target vector field is inherently time-invariant. We further introduce a time-unconditional network that eliminates explicit time-step information and infers the denoising direction solely from the spatial relationship between the current state and the noisy observation. Predicting this target vector field is equivalent to modeling the noise distribution. By avoiding overfitting to temporal trajectories, the proposed autonomous design significantly improves generation quality, robustness, and inference efficiency.

Self-supervised speech features encode both content and speaker information. Recent work introduced an SVD-based factorisation that decomposes these features into a shared content matrix capturing temporal variation and speaker-specific transformations capturing static speaker characteristics. However, how information is organised within these components remains unclear. In this paper, we investigate how the dimensions of WavLM-factorised content and speaker subspaces correlate with speech characteristics such as pitch, intensity, and voicing. We find that leading dimensions in the content space primarily capture intensity, higher-order formants, and voicing, while pitch is encoded in a later dimension. In contrast, the highest-variance speaker dimension is strongly associated with pitch and gender, with later dimensions capturing high-frequency variation. Intervention experiments show that manipulating these dimensions enables targeted control of speech characteristics for speech synthesis. Furthermore, modifying the content and speaker representations jointly provides fine-grained control over characteristics such as pitch and intensity.

More than a dozen spoken dialogue systems have recently claimed to be "full-duplex," yet the term has been used to describe substantially different capabilities. Existing surveys collapse them onto a single axis (cascaded/end-to-end, or engineered/learned) and miss the distinctions that matter most for builders. We argue that much of this ambiguity is taxonomical: current terminology does not specify where duplex decisions are made, which interaction types are supported, or how a system behaves moment by moment. This paper introduces three complementary frameworks: (i) an L0-L3 Architectural Hierarchy that locates where duplex decisions are made; (ii) a $T\times I\times R$ Interaction Ontology that specifies the temporal relation, user intent, and required system response for each interaction; and (iii) a Decision State Machine (IDLE/LISTEN/SPEAK/WAIT/DUAL) that describes how systems move between states. Across published systems and benchmarks, our audit documents a realization gap: although many architectures can in principle operate in full-duplex states, their observed behavior remains constrained by the interaction patterns represented in training and evaluation. We point to the limited public training-data coverage relative to the (largely undisclosed) industrial corpora, together with the still-unrealized goal of L3 representation-level modeling, as the key frontiers for future research on full-duplex dialogue. The related material is available at https://github.com/DuplexLM/DuplexSurvey.

Existing mean opinion score (MOS) prediction models typically predict utterance-level naturalness MOS and can be insensitive to localized pitch-accent errors. We propose Pitch-Accent-focused Speech Quality Assessment (PASQA), which explicitly targets pitch-accent correctness. To train our model, we construct a controlled Japanese accent-error dataset by changing accent patterns using an accent-controllable text-to-speech system, and compute a pseudo accent-quality score from the accent-error rate. PASQA builds on self-supervised representations and employs mora-conditioned fusion, ranking loss, an auxiliary accent-error localization task, and speaker-invariant training. Experiments show that conventional models fail to preserve the ordering by accent-error severity, whereas PASQA achieves high ordering accuracy on both seen and unseen speakers. Further, PASQA shows stronger agreement with human accent-correctness judgments. The code is available at https://github.com/lycorp-jp/PASQA.

User-defined keyword spotting (UD-KWS) enables zero-shot wake-word detection from text, but existing systems learn speaker-invariant representations that cannot reject impostors uttering the correct keyword. We address this dual zero-shot setting -- unseen keywords and unseen speakers -- with ZP-KWS, a lightweight framework combining a phoneme-supervised audio encoder with a GE2E-pretrained compact speaker encoder (about 0.9M parameters). Multiplicative late fusion at inference grants each branch independent veto power, supporting modes from conventional detection to strict speaker-gated activation without retraining. On LibriPhrase, Google Speech Commands, and Qualcomm datasets, ZP-KWS reduces target-only FRR at 1% FAR by up to 60% relative to the strongest baseline while maintaining competitive keyword detection, all within a 1.55M parameter budget for edge deployment.

Mean opinion score (MOS) prediction models are widely used as proxy metrics in text-to-speech (TTS) research, yet their ability to capture quality differences beyond acoustic fidelity remains unclear. We investigate this via controlled perturbations on speech: acoustic degradation, prosodic errors, and manipulation of speaker-specific characteristics such as pitch and speaking rate. We obtained MOS predictions for these speech samples from both human listeners and the model, and analyzed the differences in their perceptual characteristics. Results show that most models track acoustic degradation well, while all are insensitive to prosodic errors despite large subjective score drops. For speaker characteristics, models exhibit a double dissociation: strong mean fundamental frequency (F0) biases absent in human ratings, yet insensitivity to speaking rate and F0 variability that humans notice. These findings highlight limitations of scalar MOS prediction beyond acoustic fidelity.

Automatic speech recognition remains unreliable for dysarthric speech due to data scarcity and high inter-speaker variability. While synthetic data can address these gaps, traditional methods often require extensive speaker-specific data, reintroducing the collection bottleneck. We investigate zero-shot voice cloning as a low-burden augmentation strategy, using Higgs Audio V2 to clone speakers in the TORGO dataset. We fine-tune (FT) Whisper-medium on cloned, real, and hybrid data and evaluate on held-out real speech. Compared to the zero-shot (31.62%), Clone FT achieved a competitive 26.00% WER, nearly matching the 24.44% and 25.12% seen with Real and Hybrid FT, respectively. Notably, Clone and Hybrid FT outperform Real FT for moderate-severe speakers. Clone FT achieves the best results (11.45% relative) in cross-corpus evaluation on the SAP-1102. These results suggest that zero-shot cloning provides scalable training data that circumvents the costly data collection bottleneck.

Dysarthric speech recognition is crucial for facilitating effective communication among individuals with dysarthria. However, accurately recognizing dysarthric speech poses significant challenges due to varying severity levels and limited data availability. In this paper, we explore data augmentation techniques for dysarthric automatic speech recognition (ASR) systems by fine-tuning the End-to-End pre-trained Wav2Vec2 model, with a specific focus on severity levels. To address the challenges of data scarcity and the need for extensive data in fine-tuning pre-trained ASR systems for dysarthric speech, we investigate four prominent data augmentation methods: Speaking-Rate Modification (SRM), Pitch Modification (PM), Formant Modification (FM), and vocal tract Length Perturbation (VTLP), tailored to different aspects of dysarthria. The study uses individually fine-tuned Wav2Vec2 models for each severity class as baseline systems. Additionally, we conducted severity-specific fine-tuning of the ASR model using augmented data. Results demonstrate distinct efficacy patterns for each augmentation technique across severity levels. The best WERs were achieved with SRM ($s$=0.8) for \textit{low} (9.02\%) and \textit{medium} (38.11\%) severities, and with PM ($τ$=0.8) for \textit{high} severity (55.15\%), reflecting relative improvements of 30.02\%, 16.64\%, and 15.47\%, respectively. These results confirm the effectiveness of the augmentation methods in improving dysarthric ASR performance.

The challenge associated with recognizing dysarthric speech primarily arises from pronounced acoustic variability attributed to impaired articulatory precision. Past research has demonstrated improved recognition through the use of hybrid DNN/HMM sequence discriminative training. This paper presents a comprehensive investigation of various combinations of acoustic features tailored to different Acoustic Models, offering suitable feature selections for each. The incorporation of Pitch features notably improved recognition performance, especially for sentence recognition tasks involving dysarthric speech. Through a systematic examination of the TORGO database, we have demonstrated the potential to enhance the performance of the state-of-the-art Factorized Time Delay Neural Network (F-TDNN) model for recognizing dysarthric speech. Our methods, implemented with the F-TDNN model, resulted in a 4.65\% relative improvement in isolated word recognition and a 4.63\% relative improvement in sentence recognition for dysarthric speech, compared to previous research. This improvement effectively compensates for speech variability, attributable to our deliberate selection of the number of overlapping frames between consecutive training example chunks.

The challenge associated with recognizing dysarthric speech primarily arises from pronounced acoustic variability attributed to impaired articulatory precision. Past research has demonstrated improved recognition through the use of hybrid DNN/HMM sequence discriminative training. This paper presents a comprehensive investigation of various combinations of acoustic features tailored to different Acoustic Models, offering suitable feature selections for each. The incorporation of Pitch features notably improved recognition performance, especially for sentence recognition tasks involving dysarthric speech. Through a systematic examination of the TORGO database, we have demonstrated the potential to enhance the performance of the state-of-the-art Factorized Time Delay Neural Network (F-TDNN) model for recognizing dysarthric speech. Our methods, implemented with the F-TDNN model, resulted in a 4.65\% relative improvement in isolated word recognition and a 4.63\% relative improvement in sentence recognition for dysarthric speech, compared to previous research. This improvement effectively compensates for speech variability, attributable to our deliberate selection of the number of overlapping frames between consecutive training example chunks.

Unified Multimodal Modeling aims to integrate visual understanding and generation within a single system. However, existing approaches typically rely on two disparate visual tokenizers, which splits the representation space and hinders truly unified modeling. We propose UniAR, a unified autoregressive framework where a single discrete visual tokenizer serves as the key bridge between understanding and generation, enabling a shared context in which the model can directly interpret its own generated visual tokens without additional re-encoding. UniAR adapts a pretrained vision encoder with multi-level feature fusion and a lookup-free bitwise quantization scheme, preserving both high-level semantics and low-level details while scaling the effective visual vocabulary at minimal cost. Building on this, the unified autoregressive model adopts parallel-bitwise-prediction to jointly predict spatially grouped, multi-level visual codes, substantially reducing visual sequence length and accelerating generation. Finally, a diffusion-based visual decoder operates on discrete visual tokens to decode high-fidelity images. Through large-scale pre-training, followed by supervised fine-tuning and reinforcement learning, UniAR achieves state-of-the-art performance on image generation and image editing while remaining competitive on multimodal understanding benchmarks. The project page is available at https://sharelab-sii.github.io/uniar-web.

We present TerraMind, the first any-to-any generative, multimodal foundation model for Earth observation (EO). Unlike other multimodal models, TerraMind is pretrained on dual-scale representations combining both token-level and pixel-level data across modalities. On a token level, TerraMind encodes high-level contextual information to learn cross-modal relationships, while on a pixel level, TerraMind leverages fine-grained representations to capture critical spatial nuances. We pretrained TerraMind on nine geospatial modalities of a global, large-scale dataset. In this paper, we demonstrate that (i) TerraMind's dual-scale early fusion approach unlocks a range of zero-shot and few-shot applications for Earth observation, (ii) TerraMind introduces "Thinking-in-Modalities" (TiM) -- the capability of generating additional artificial data during finetuning and inference to improve the model output -- and (iii) TerraMind achieves beyond state-of-the-art performance in community-standard benchmarks for EO like PANGAEA. The pretraining dataset, the model weights, and our code are open-sourced under a permissive license.

Multimodal large language models (MLLMs) have achieved remarkable progress in visual understanding tasks. However, most existing MLLMs rely on autoregressive generation, which limits their efficiency for perception tasks that require captioning multiple regions. In this work, we propose PerceptionDLM, a multimodal diffusion language model optimized for efficient parallel region perception. Built upon PerceptionDLM-Base, a strong foundational baseline that achieves state-of-the-art performance among open-source diffusion MLLMs, our architecture fully leverages the parallel decoding nature of DLMs. Specifically, we introduce efficient prompting and structured attention masking to enable simultaneous perception of multiple masked regions, allowing the model to generate region descriptions in parallel at both the sequence and token levels. This design significantly improves inference efficiency compared with existing approaches that process regions sequentially. To systematically evaluate the parallelism property of visual perception capability for DLMs, we construct a new Parallel Detailed Localized Captioning Benchmark (ParaDLC-Bench) by scaling the DLC-Bench to include multiple region masks per image, enabling joint evaluation of both caption quality and inference efficiency. Experiments demonstrate that PerceptionDLM maintains competitive performance in region captioning while achieving substantial speed improvements for multi-region perception tasks. Our results highlight the potential of multimodal diffusion language models for efficient, parallel visual perception. To the best of our knowledge, we are the first to achieve parallel region caption and perception by leveraging the advantages of diffusion language models. Code, models, and datasets are released.

Multimodal Large Language Models (MLLMs) excel at 2D semantic understanding but lack intrinsic 3D awareness, resulting in representations that fail to maintain geometric and spatial consistency across video frames. Given the scarcity of large-scale 3D data, we present GeoVR, a novel framework that learns geometric representations using purely 2D video sequences. This approach effectively restructures the semantic latent space within MLLMs to unlock spatial intelligence. Rather than employing superficial feature mixing, GeoVR reshapes the internal representations of the MLLM by distilling geometry knowledge from pre-trained 3D foundation models. This is accomplished through a multi-objective learning strategy driven by four complementary geometric targets: (1) estimating inter-frame camera poses to embed varying viewpoint dynamics, (2) regressing dense depth maps to anchor physical distances, (3) predicting a metric scale factor for real-world calibration, and (4) distilling multi-scale 3D features to align the intermediate feature space. Guided by these explicit physical and geometric constraints, the model's internal representations naturally develop strong 3D awareness. Extensive experiments on spatial reasoning benchmarks demonstrate that GeoVR achieves state-of-the-art performance, establishing a new paradigm for endowing foundation models with spatial intelligence.

Recent progress in vision-language models has enabled the processing of increasingly long video sequences, but the ability to handle extended token streams does not translate to understanding of narrative structure in long videos. Existing long video benchmarks focus on needle-in-a-haystack retrieval rather than evaluating how low-level actions form events, how events interact across time, and how narratives progress, for example, whether a model can connect an early setback, such as a job loss to a later relationship breakup, despite long gaps, intervening scenes, or flashbacks that reframe what occurred. We introduce NEST (Narrative Event Structures in Time for Long Video Understanding), a dataset of 1005 full-length movies (avg. 98 minutes), each annotated with 102 multimodal narrative events grounded in visual content, dialogue, and audio. NEST captures multimodal narrative events with structured annotations grounded in visual content, dialogue, and audio, and links them through relations that reflect narrative structure, including temporal ordering, hierarchical composition, and long-range dependencies. We introduce baselines for event trigger detection (ETD), event localization (EL), event argument extraction (EAE), and event relation extraction (ERE). The benchmark is highly challenging for grounded event discovery, with ETD below 8%, EL under 6%, and EAE below 11%. In contrast, ERE is more tractable once events are given, reaching 35.45% F1 zero-shot and 44.42% F1 after fine-tuning.

Multimodal time series forecasting, which pairs numerical sequences with domain-relevant textual reports, promises to inject world knowledge into forecasting pipelines. However, we uncover a critical failure mode in existing frameworks that we term text collapse: the text branch converges to a content-independent transformation, contributing negligible discriminative signal regardless of the input description. We argue that text collapse is a consequence of a fundamental asymmetry in time series forecasting: the numerical input is strongly autocorrelated with the output, making the numerical backbone inherently dominant, while the text branch, despite carrying complementary and often critical information, is insufficiently utilized, leading to its systematic underexploitation. To address this, we propose \textbf{REST-TS} (\textbf{R}esidual-\textbf{E}xclusive \textbf{S}upervision for \textbf{T}ext in \textbf{T}ime \textbf{S}eries), which turns the asymmetry into a design principle: the numerical backbone produces its own independent numerical forecast, and the text branch is exclusively supervised to predict the structured components of the residual, the prediction gap that numbers cannot explain. Because no numerical pathway can reduce these losses, the text branch must extract genuine content from the input description. Evaluated across diverse real-world domains and backbone architectures, REST-TS achieves state-of-the-art performance and consistently demonstrates greater text-branch utilization than existing frameworks, providing strong empirical evidence that supervising the text branch on the residual compels it to extract genuine content from the input.

Multimodal large language models (MLLMs) are increasingly deployed in personally and societally consequential settings, yet the visual cues that shape how these models judge people remain poorly understood. Prior work often compares different (groups of) individuals, making it difficult to separate appearance effects from identity differences. We introduce StylisticBias, a controlled benchmark for evaluating attribute-level social bias in MLLMs. We generate 500 photorealistic base faces and create about 50 single-attribute variations per face, producing about 25K images. This design keeps identity fixed and changes one visual attribute at a time. It lets us measure how specific cues shift model judgments. We evaluate six MLLMs across 25 binary social judgment scenarios. We find that age and body type dominate identity-level effects, while fashion style and other visual cues drive the largest attribute-level shifts. We further find that about 15 attributes account for nearly 80\% of the total variation, showing that bias is concentrated in a small set of visual cues. Sensitivity is strongest in judgments that are semantically aligned with appearance, especially socioeconomic and style-related judgments. We release StylisticBias as a benchmark for fine-grained bias evaluation in multimodal models. Code and dataset: https://github.com/timo-cavelius/StylisticBias and https://hf.co/datasets/shaghayegh/stylistic-bias-dataset.

Concept Bottleneck Models (CBMs) enhance the interpretability of deep learning networks by aligning the features extracted from images with natural concepts. However, existing CBMs are constrained in their ability to generalize beyond a fixed set of predefined classes and the risk of non-concept information leakage, where predictive signals outside the intended concepts are inadvertently exploited. In this paper, we propose Multimodal Concept Bottleneck Model (MM-CBM) to address these issues and extend CBMs into CLIP. MM-CBM utilizes dual Concept Bottleneck Layers (CBLs) to align both the image and text embeddings into interpretable features. This allows us to perform new vision tasks like zero-shot classification or image retrieval in an interpretable way. Compared to existing methods, MM-CBM achieves up to 51.26% accuracy improvement on average across four standard benchmarks. Our method maintains high accuracy, staying within ~5% of black-box performance while offering greater interpretability.

Language models have become essential tools in shaping modern workflows. However, their global effectiveness hinges on a nuanced understanding of local socio-cultural contexts. To address this gap, we present NRITYAM, a comprehensive benchmark for evaluating the cultural comprehension capabilities of language models in the context of global dance traditions. NRITYAM comprises 9,260 carefully curated question-answer pairs spanning 12 languages, making it the largest dataset dedicated to evaluating cultural knowledge in dance. The dataset has been developed from the ground up through close collaboration with native dance artists and native speakers of the languages, who authored and validated culturally relevant questions specific to their regions. We evaluate a broad set of models, including large language models, small language models, multimodal large language models, and small multimodal language models. As a multilingual and multicultural benchmark, NRITYAM sets a new standard for evaluating the ability of AI systems to understand and reason about traditional performing arts. Detailed dataset samples are available at~\url{https://github.com/niladrighosh03/NRITYAM}.

Recent advances in multimodal foundation models unifying image understanding and generation have opened exciting avenues for tackling a wide range of vision-language tasks within a single framework. Despite progress, existing unified models typically require extensive pretraining and struggle to achieve the same level of performance compared to models dedicated to each task. Additionally, many of these models suffer from slow image generation speeds, limiting their practical deployment in real-time or resource-constrained settings. In this work, we propose Layerwise Timestep-Expert Flow-based Transformer (LaTtE-Flow), a novel and efficient architecture that unifies image understanding and generation within a single multimodal model. LaTtE-Flow builds upon powerful pretrained Vision-Language Models (VLMs) to inherit strong multimodal understanding capabilities, and extends them with a novel Layerwise Timestep Experts flow-based architecture for efficient image generation. LaTtE-Flow distributes the flow-matching process across specialized groups of Transformer layers, each responsible for a distinct subset of timesteps. This design significantly improves sampling efficiency by activating only a small subset of layers at each sampling timestep. To further enhance performance, we propose a Timestep-Conditioned Residual Attention mechanism for efficient information reuse across layers. Experiments demonstrate that LaTtE-Flow achieves strong performance on multimodal understanding tasks, while achieving competitive image generation quality with around 6x faster inference speed compared to recent unified multimodal models.

Leveraging Multimodal Large Language Models (MLLMs) via contrastive learning has become a mainstream paradigm for improving the performance of Universal Multimodal Retrieval (UMR). However, previous works have ignored the grain blindness when adapting the contrastive paradigm into retrieval tasks. Grain blindness refers to the tendency of the model to overlook grain-level information contained in the query, which is crucial for effectively handling complex queries. This stems from contrastive learning treating samples as a binary classification (positive/negative), while ignoring the different information carried by each negative sample. To address this, we argue that negatives should be treated differently according to their similarity to the positive sample, enabling the model to learn distinct grain information from each negative. In this paper, we introduce a simple but effective framework, called ELVA, a novel rule-based RL framework that mitigates grain blindness through ranking-driven MLLMs. 1) Instead of relying on reward models, we extend Reinforcement Learning with Verifiable Rewards (RLVR) to retrieval tasks, allowing the model to explore new ranking behaviors without explicit ranking labels. 2) By utilizing rule-based rewards, our approach jointly optimizes the ranking of negative samples while enlarging the similarity gap between positive and negative. To more precisely measure grain blindness, we further introduce MRBench, a new benchmark specifically designed for multi-grain query scenarios. ELVA achieves state-of-the-art results across standard retrieval benchmarks, and its notable 13.1% improvement on MRBench further demonstrates its effectiveness in alleviating grain blindness.

Multimodal foundation models have advanced rapidly thanks to large optical benchmarks, but comparable resources for synthetic aperture radar (SAR) remain limited. Existing SAR--optical datasets largely rely on low-resolution, intensity-only Ground Range Detected~(GRD) products and do not preserve complex-valued SAR measurements or native acquisition geometry, which restricts physically grounded multimodal learning. In particular, large-scale public datasets combining very-high-resolution (VHR) SAR SLC, aligned optical imagery, and natural-language descriptions are still lacking. We present a VHR SAR--optical--text dataset built from open-access Umbra spotlight acquisitions distributed as Sensor Independent Complex Data (SICD). From around 2,500 worldwide scenes (VV/HH, 20cm--2m native resolution), we standardize all SAR data to an 80cm slant-range grid via band-limited FFT resampling and tile the imagery into 1024 by 1024 patches. For each SAR patch, we retrieve a high-resolution optical tile and warp it into the SAR grid using local coordinate correspondences for local pixel-level alignment. We further generate three caption variants (SHORT/MID/LONG) per sample to support vision--language training and evaluation. Our dataset contains 119,566 triplets (complex and amplitude slant-range SAR patch, aligned optical patch, natural-language description) covering 257 locations across 72 countries and a broad range of land types and infrastructures. We release fixed train/validation/test splits and the full preprocessing and baseline code to enable reproducible benchmarks for multimodal alignment on cross-modal retrieval and conditional generation in native SAR geometry. The dataset is publicly available on the Hugging Face Hub at https://huggingface.co/datasets/ONERA/SARLO-80.