SVHalluc benchmark shows open-source audio-visual LLMs achieve near-random accuracy on semantic and temporal speech-vision alignment tasks while Gemini 2.5 Pro performs substantially better.
super hub Mixed citations
Qwen2.5-Omni Technical Report
Mixed citation behavior. Most common role is background (59%).
225
Pith papers citing it
8
external citations · Pith
Background
59% of classified citations
abstract
In this report, we present Qwen2.5-Omni, an end-to-end multimodal model designed to perceive diverse modalities, including text, images, audio, and video, while simultaneously generating text and natural speech responses in a streaming manner. To enable the streaming of multimodal information inputs, both audio and visual encoders utilize a block-wise processing approach. To synchronize the timestamps of video inputs with audio, we organize the audio and video sequentially in an interleaved manner and propose a novel position embedding approach, named TMRoPE(Time-aligned Multimodal RoPE). To concurrently generate text and speech while avoiding interference between the two modalities, we propose \textbf{Thinker-Talker} architecture. In this framework, Thinker functions as a large language model tasked with text generation, while Talker is a dual-track autoregressive model that directly utilizes the hidden representations from the Thinker to produce audio tokens as output. Both the Thinker and Talker models are designed to be trained and inferred in an end-to-end manner. For decoding audio tokens in a streaming manner, we introduce a sliding-window DiT that restricts the receptive field, aiming to reduce the initial package delay. Qwen2.5-Omni is comparable with the similarly sized Qwen2.5-VL and outperforms Qwen2-Audio. Furthermore, Qwen2.5-Omni achieves state-of-the-art performance on multimodal benchmarks like Omni-Bench. Notably, Qwen2.5-Omni's performance in end-to-end speech instruction following is comparable to its capabilities with text inputs, as evidenced by benchmarks such as MMLU and GSM8K. As for speech generation, Qwen2.5-Omni's streaming Talker outperforms most existing streaming and non-streaming alternatives in robustness and naturalness.
hub tools
citation-role summary
background 20
baseline 10
method 2
citation-polarity summary
claims ledger
- abstract In this report, we present Qwen2.5-Omni, an end-to-end multimodal model designed to perceive diverse modalities, including text, images, audio, and video, while simultaneously generating text and natural speech responses in a streaming manner. To enable the streaming of multimodal information inputs, both audio and visual encoders utilize a block-wise processing approach. To synchronize the timestamps of video inputs with audio, we organize the audio and video sequentially in an interleaved manner and propose a novel position embedding approach, named TMRoPE(Time-aligned Multimodal RoPE). To c
authors
co-cited works
representative citing papers
Omni-DeepSearch is a 640-sample benchmark for audio-driven omni-modal search where the best model reaches only 43.44% accuracy, exposing bottlenecks in audio inference, tool use, and cross-modal reasoning.
TraceAV-Bench is the first benchmark for multi-hop trajectory reasoning over long audio-visual videos, showing top models reach only 51-68% accuracy with substantial room for improvement.
HalluAudio is the first large-scale benchmark spanning speech, environmental sound, and music that uses human-verified QA pairs, adversarial prompts, and mixed-audio tests to measure hallucinations in large audio-language models.
Benign fine-tuning on audio data breaks safety alignment in Audio LLMs by raising jailbreak success rates up to 87%, with the dominant risk axis depending on model architecture and embedding proximity to harmful content.
DialBGM is a new benchmark dataset revealing that existing AI models fall far short of human performance when recommending fitting background music for open-domain conversations.
AVLLMs encode audio semantics in middle layers but suppress them in final text outputs when audio conflicts with vision, due to training that largely inherits from vision-language base models.
EgoSound is a new benchmark with 7315 QA pairs across seven tasks to evaluate egocentric sound understanding in multimodal large language models.
Cornfigurator is the first automated deployment planner for generic any-to-any multimodal models that explores the full range of colocation-to-disaggregation strategies and delivers 1.12x to 6.32x higher goodput than existing systems or expert plans.
A two-stage replay-based post-training method corrects ASR timestamp drift across non-speech gaps while preserving recognition far better than ordinary timestamp fine-tuning.
SpeechCombine produces instruction-following SLMs via speech pre-training followed by direct weight combination with the text LLM instruction delta, without any speech instruction tuning.
FlexiSLM is the first spoken language model supporting dynamic and controllable frame rates on speech input and output, outperforming fixed-rate 7B models at high quality and enabling faster inference at lower rates like 6.25 Hz.
MuseBench shows state-of-the-art MLLMs achieve only 48.29% accuracy on intent-level audiovisual arts understanding versus 87.18% for human experts.
PreferenceASR is a preference-aware ASR test set built from seven corpora that shows model rankings change when user output-style instructions are considered.
Introduces CASU benchmark with four tasks to evaluate context-aware auditory scene understanding in LALMs via semi-synthetic audio compositions of speech, events, and environments.
ParaPairAudioBench is a new pairwise benchmark showing LALM judges lag human paralinguistic judgments by 32 percentage points with poor tie calibration across style, rate, emphasis, age, and gender.
video-SALMONN-R³ is an end-to-end video-LLM trained with reinforcement learning to perform selective re-watching, re-asking, and re-answering for efficient video question answering without chain-of-thought cold-start supervision.
AudioCALM presents a continuous autoregressive framework with flow-matching prediction and A-MoME architecture that unifies speech, sound, and music generation while matching modality-specific state-of-the-art performance.
A survey proposing an L0-L3 architectural hierarchy, T×I×R interaction ontology, and IDLE/LISTEN/SPEAK/WAIT/DUAL decision state machine for full-duplex spoken dialogue systems, documenting a realization gap between architectural potential and observed behavior due to training data limits.
M* introduces the Walk Graph abstraction to serve arbitrary compositions of multimodal model components and reports latency and throughput gains over vLLM-Omni and other baselines on text-to-image, text-to-speech, and robotic planning workloads.
AVLLMs route audio-visual information sequentially in video tasks and via parallel streams for interleaved items, allowing early token discard with little performance loss across models and scales.
MLLMs fail to detect absent correct answers in video QA tasks across three evaluation settings, defaulting to distractors even with chain-of-thought prompting.
Phun-Bench shows LLMs recall pronunciations well but struggle to apply phonological knowledge flexibly like humans.
UniSinger unifies speaker-cloned song generation and accompaniment co-generation SVC in one multimodal diffusion transformer model trained with curriculum learning via task-specific modality masking.
citing papers explorer
-
SVHalluc: Benchmarking Speech-Vision Hallucination in Audio-Visual Large Language Models
SVHalluc benchmark shows open-source audio-visual LLMs achieve near-random accuracy on semantic and temporal speech-vision alignment tasks while Gemini 2.5 Pro performs substantially better.
-
Omni-DeepSearch: A Benchmark for Audio-Driven Omni-Modal Deep Search
Omni-DeepSearch is a 640-sample benchmark for audio-driven omni-modal search where the best model reaches only 43.44% accuracy, exposing bottlenecks in audio inference, tool use, and cross-modal reasoning.
-
TraceAV-Bench: Benchmarking Multi-Hop Trajectory Reasoning over Long Audio-Visual Videos
TraceAV-Bench is the first benchmark for multi-hop trajectory reasoning over long audio-visual videos, showing top models reach only 51-68% accuracy with substantial room for improvement.
-
HalluAudio: A Comprehensive Benchmark for Hallucination Detection in Large Audio-Language Models
HalluAudio is the first large-scale benchmark spanning speech, environmental sound, and music that uses human-verified QA pairs, adversarial prompts, and mixed-audio tests to measure hallucinations in large audio-language models.
-
Benign Fine-Tuning Breaks Safety Alignment in Audio LLMs
Benign fine-tuning on audio data breaks safety alignment in Audio LLMs by raising jailbreak success rates up to 87%, with the dominant risk axis depending on model architecture and embedding proximity to harmful content.
-
DialBGM: A Benchmark for Background Music Recommendation from Everyday Multi-Turn Dialogues
DialBGM is a new benchmark dataset revealing that existing AI models fall far short of human performance when recommending fitting background music for open-domain conversations.
-
Do Audio-Visual Large Language Models Really See and Hear?
AVLLMs encode audio semantics in middle layers but suppress them in final text outputs when audio conflicts with vision, due to training that largely inherits from vision-language base models.
-
EgoSound: Benchmarking Sound Understanding in Egocentric Videos
EgoSound is a new benchmark with 7315 QA pairs across seven tasks to evaluate egocentric sound understanding in multimodal large language models.
-
Cornfigurator: Automated Planning for Any-to-Any Multimodal Model Serving
Cornfigurator is the first automated deployment planner for generic any-to-any multimodal models that explores the full range of colocation-to-disaggregation strategies and delivers 1.12x to 6.32x higher goodput than existing systems or expert plans.
-
REDDIT: Correcting Model-Generated Timestamp Drift in ASR without Forgetting via Replay-Based Distribution Editing
A two-stage replay-based post-training method corrects ASR timestamp drift across non-speech gaps while preserving recognition far better than ordinary timestamp fine-tuning.
-
Unlocking Speech-Text Compositional Powers: Instruction-Following Speech Language Models without Instruction Tuning
SpeechCombine produces instruction-following SLMs via speech pre-training followed by direct weight combination with the text LLM instruction delta, without any speech instruction tuning.
-
FlexiSLM: A Dynamic and Controllable Frame Rate Spoken Language Model
FlexiSLM is the first spoken language model supporting dynamic and controllable frame rates on speech input and output, outperforming fixed-rate 7B models at high quality and enabling faster inference at lower rates like 6.25 Hz.
-
MuseBench: Benchmarking Intent-Level Audiovisual Arts Understanding in MLLMs
MuseBench shows state-of-the-art MLLMs achieve only 48.29% accuracy on intent-level audiovisual arts understanding versus 87.18% for human experts.
-
Preference-ASR: A Preference-Aware Test Set for Benchmarking ASR in the Era of Speech LLMs
PreferenceASR is a preference-aware ASR test set built from seven corpora that shows model rankings change when user output-style instructions are considered.
-
From Sounds to Scenes: A Benchmark for Evaluating Context-Aware Auditory Scene Understanding in Large Audio Language Models
Introduces CASU benchmark with four tasks to evaluate context-aware auditory scene understanding in LALMs via semi-synthetic audio compositions of speech, events, and environments.
-
ParaPairAudioBench: Paralinguistic Pairwise Audio Benchmark for LALM-as-a-Judge
ParaPairAudioBench is a new pairwise benchmark showing LALM judges lag human paralinguistic judgments by 32 percentage points with poor tie calibration across style, rate, emphasis, age, and gender.
-
video-SALMONN-R$^3$: Learning to ReWatch, ReAsk, and ReAnswer for Efficient Video Understanding
video-SALMONN-R³ is an end-to-end video-LLM trained with reinforcement learning to perform selective re-watching, re-asking, and re-answering for efficient video question answering without chain-of-thought cold-start supervision.
-
AudioCALM: Continuous Autoregressive Language Modeling for Universal Audio Generation
AudioCALM presents a continuous autoregressive framework with flow-matching prediction and A-MoME architecture that unifies speech, sound, and music generation while matching modality-specific state-of-the-art performance.
-
A Survey of Full-Duplex Spoken Dialogue Systems: Architectural Hierarchy, Interaction Ontology, and Decision State Machine
A survey proposing an L0-L3 architectural hierarchy, T×I×R interaction ontology, and IDLE/LISTEN/SPEAK/WAIT/DUAL decision state machine for full-duplex spoken dialogue systems, documenting a realization gap between architectural potential and observed behavior due to training data limits.
-
M*: A Modular, Extensible, Serving System for Multimodal Models
M* introduces the Walk Graph abstraction to serve arbitrary compositions of multimodal model components and reports latency and throughput gains over vLLM-Omni and other baselines on text-to-image, text-to-speech, and robotic planning workloads.
-
From Senses to Decisions: The Information Flow of Auditory and Visual Perception in Multimodal LLMs
AVLLMs route audio-visual information sequentially in video tasks and via parallel streams for interleaved items, allowing early token discard with little performance loss across models and scales.
-
When No Answer Is Correct: Diagnosing Absent Answer Detection for MLLMs in Video Understanding
MLLMs fail to detect absent correct answers in video QA tasks across three evaluation settings, defaulting to distractors even with chain-of-thought prompting.
-
Phun-Bench: Evaluating LLMs on Phonological Understanding in Chinese
Phun-Bench shows LLMs recall pronunciations well but struggle to apply phonological knowledge flexibly like humans.
-
Towards Unified Song Generation and Singing Voice Conversion with Accompaniment Co-Generation
UniSinger unifies speaker-cloned song generation and accompaniment co-generation SVC in one multimodal diffusion transformer model trained with curriculum learning via task-specific modality masking.
-
AVI-Bench: Toward Human-like Audio-Visual Intelligence of Omni-MLLMs
AVI-Bench is a cognitively inspired benchmark that evaluates Omni-MLLMs on joint audio-visual tasks and reveals substantial limitations in current models.
-
X-Stream: Exploring MLLMs as Multiplexers for Multi-Stream Understanding
X-Stream benchmark shows SOTA MLLMs score ~50% on concurrent multi-stream tasks and lack proactive ability, using a dual-verification pipeline to avoid single-stream bias.
-
PolySpeech-100: A Large-Scale Benchmark for Speech Understanding Across 100+ Languages and Dialects
PolySpeech-100 is a new benchmark for native-level speech comprehension across 110 linguistic variants that evaluates 22 models and reports E2E advantages on dialects, robustness gaps on low-resource languages, and degradation from Chain-of-Thought prompting.
-
Sandboxed Coding Agents are Competitive Omni-modal Task Solvers
Sandboxed coding agents with text+image access match or outperform native omnimodal models on audio-video benchmarks by converting tasks into code-driven retrieval and processing.
-
VideoFDB: Evaluating Full-Duplex Vision-Speech Capabilities in Conversational Agents
Introduces the first full-duplex audio-visual-to-audio-visual benchmark, showing current agents fail to use visual nonverbal cues in live conversation.
-
Native Audio-Visual Alignment for Generation
NAVA proposes native audio-visual alignment via Align-then-Fuse MMDiT and Timbre-in-Context Conditioning for joint audio-video generation with improved synchronization and timbre control.
-
AVBench: Human-Aligned and Automated Evaluation Benchmark for Audio-Video Generative Models
AVBench is a benchmark for human-centric AV generation evaluation featuring ten fine-grained dimensions and preference-learned evaluators that output continuous probabilistic scores from binary decisions.
-
LatentOmni: Rethinking Omni-Modal Understanding via Unified Audio-Visual Latent Reasoning
LatentOmni proposes a latent-space cross-modal reasoning framework that uses feature-level supervision and Omni-Sync Position Embedding to align and synchronize audio-visual latents, supported by a new 35K interleaved reasoning dataset and showing gains over text CoT baselines.
-
WikiVQABench: A Knowledge-Grounded Visual Question Answering Benchmark from Wikipedia and Wikidata
WikiVQABench is a human-curated collection of Wikipedia-based VQA items that require both visual evidence and external knowledge from Wikidata to answer correctly.
-
Codec-Robust Attacks on Audio LLMs
CodecAttack perturbs audio in codec latent space with multi-bitrate EoT to achieve 85.5% average ASR on Opus-compressed Audio LLMs versus under 26% for waveform baselines, with transfer to MP3 and AAC.
-
AffectVerse: Emotional World Models for Multimodal Affective Computing
AffectVerse improves multimodal emotion recognition by at least 2.57% on nine benchmarks through an Emotion World Module that performs short-horizon latent affective prediction via cross-modal temporal imagination and belief aggregation.
-
OmniPro: A Comprehensive Benchmark for Omni-Proactive Streaming Video Understanding
OmniPro is the first benchmark jointly evaluating omni-modal perception, proactive responding, and diverse streaming video understanding tasks using a dual-mode protocol on 2700 samples.
-
CBT-Audio: Evaluating Audio Language Models for Patient-Side Distress Intensity Estimation in CBT Session Recordings
CBT-Audio dataset shows that adding audio input improves distress intensity estimation over transcripts alone for 8 of 10 audio language models, with clearest gains when verbal content and vocal delivery diverge.
-
Boosting Omni-Modal Language Models: Staged Post-Training with Visually Debiased Evaluation
Visual debiasing of omni-modal benchmarks combined with staged post-training lets a 3B model match or exceed a 30B model without a stronger teacher.
-
AffectCodec: Emotion-Preserving Neural Speech Codec for Expressive Speech Modeling
AffectCodec is an emotion-guided neural speech codec that preserves emotional cues during quantization while maintaining semantic fidelity and prosodic naturalness.
-
How Should LLMs Listen While Speaking? A Study of User-Stream Routing in Full-Duplex Spoken Dialogue
Channel fusion gives better semantic grounding and QA performance in full-duplex LLM dialogue but is vulnerable to context corruption during interruptions, while cross-attention routing is more robust at the cost of weaker integration.
-
VITA-QinYu: Expressive Spoken Language Model for Role-Playing and Singing
VITA-QinYu is the first expressive end-to-end spoken language model supporting role-playing and singing alongside conversation, trained on 15.8K hours of data and outperforming prior models on expressiveness and conversational benchmarks.
-
Sparse Tokens Suffice: Jailbreaking Audio Language Models via Token-Aware Gradient Optimization
TAGO performs sparse jailbreak optimization on audio LMs by retaining only high-gradient-energy tokens, preserving near-full ASR at 25% retention across three models.
-
TMD-Bench: A Multi-Level Evaluation Paradigm for Music-Dance Co-Generation
TMD-Bench is a multi-level benchmark that measures music-dance co-generation quality including beat-level rhythmic synchronization, supported by a new dataset and Music Captioner, and shows commercial models lag in rhythm while a new baseline performs competitively.
-
Walking Through Uncertainty: An Empirical Study of Uncertainty Estimation for Audio-Aware Large Language Models
Semantic-level and verification-based uncertainty methods outperform token-level baselines for audio reasoning in ALLMs, but their relative performance on hallucination and unanswerable-question benchmarks is model- and task-dependent.
-
Toward Efficient Membership Inference Attacks against Federated Large Language Models: A Projection Residual Approach
ProjRes achieves near-100% accuracy in membership inference on FedLLMs by measuring projection residuals of hidden embeddings on gradient subspaces, outperforming prior methods by up to 75.75% even under differential privacy.
-
SpeechParaling-Bench: A Comprehensive Benchmark for Paralinguistic-Aware Speech Generation
SpeechParaling-Bench is a new evaluation framework for paralinguistic-aware speech generation that reveals major limitations in current large audio-language models.
-
ATIR: Towards Audio-Text Interleaved Contextual Retrieval
Defines ATIR task and benchmark for mixed audio-text queries; MLLM model with token compression shows substantial gains over strong baselines.
-
ProjLens: Unveiling the Role of Projectors in Multimodal Model Safety
ProjLens shows that backdoor parameters in MLLMs are encoded in low-rank subspaces of the projector and that embeddings shift toward the target direction with magnitude linear in input norm, activating only on poisoned samples.
-
Watching Movies Like a Human: Egocentric Emotion Understanding for Embodied Companions
Creates the first egocentric screen-view movie emotion benchmark and demonstrates that cinematic models drop sharply in Macro-F1 on realistic robot-like viewing conditions while domain-specific training improves robustness.
-
From Reactive to Proactive: Assessing the Proactivity of Voice Agents via ProVoice-Bench
ProVoice-Bench is the first framework to evaluate proactive voice agents, revealing that state-of-the-art multimodal LLMs struggle with over-triggering and context-aware reasoning.