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.
super hub Canonical reference
Qwen2-Audio Technical Report
Canonical reference. 76% of citing Pith papers cite this work as background.
122
Pith papers citing it
Background
76% of classified citations
abstract
We introduce the latest progress of Qwen-Audio, a large-scale audio-language model called Qwen2-Audio, which is capable of accepting various audio signal inputs and performing audio analysis or direct textual responses with regard to speech instructions. In contrast to complex hierarchical tags, we have simplified the pre-training process by utilizing natural language prompts for different data and tasks, and have further expanded the data volume. We have boosted the instruction-following capability of Qwen2-Audio and implemented two distinct audio interaction modes for voice chat and audio analysis. In the voice chat mode, users can freely engage in voice interactions with Qwen2-Audio without text input. In the audio analysis mode, users could provide audio and text instructions for analysis during the interaction. Note that we do not use any system prompts to switch between voice chat and audio analysis modes. Qwen2-Audio is capable of intelligently comprehending the content within audio and following voice commands to respond appropriately. For instance, in an audio segment that simultaneously contains sounds, multi-speaker conversations, and a voice command, Qwen2-Audio can directly understand the command and provide an interpretation and response to the audio. Additionally, DPO has optimized the model's performance in terms of factuality and adherence to desired behavior. According to the evaluation results from AIR-Bench, Qwen2-Audio outperformed previous SOTAs, such as Gemini-1.5-pro, in tests focused on audio-centric instruction-following capabilities. Qwen2-Audio is open-sourced with the aim of fostering the advancement of the multi-modal language community.
hub tools
citation-role summary
background 19
baseline 5
method 1
citation-polarity summary
claims ledger
- abstract We introduce the latest progress of Qwen-Audio, a large-scale audio-language model called Qwen2-Audio, which is capable of accepting various audio signal inputs and performing audio analysis or direct textual responses with regard to speech instructions. In contrast to complex hierarchical tags, we have simplified the pre-training process by utilizing natural language prompts for different data and tasks, and have further expanded the data volume. We have boosted the instruction-following capability of Qwen2-Audio and implemented two distinct audio interaction modes for voice chat and audio an
authors
co-cited works
representative citing papers
ReasonAudio benchmark reveals that state-of-the-art text-audio retrieval models struggle with reasoning tasks like negation and duration, and multimodal LLMs lose reasoning ability after contrastive fine-tuning.
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.
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.
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.
MusTBENCH evaluates temporal grounding in large audio-language models via five expert-validated tasks, and MusT improves performance through encoder adaptation, LLM adaptation, supervised fine-tuning, and RL optimization.
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.
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.
ToxiAlert-Bench dataset and dual-head neural network detect toxic speech by distinguishing textual versus paralinguistic sources, reporting 21.1% Macro-F1 and 13% accuracy gains over baselines.
SpurAudio benchmark shows state-of-the-art few-shot audio classifiers suffer large performance drops when background correlations are disrupted, even in large pretrained models.
NAACA uses a neuro-inspired oscillatory working memory to gate attention in audio language models, raising AudioQwen's average precision from 53.5% to 70.6% on XD-Violence while cutting unnecessary calls.
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.
MIST is a new synthetic speech-based tool-calling dataset for IoT devices that exposes performance gaps between open- and closed-weight multimodal LLMs.
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.
Introduces the Indic-CodecFake dataset for Indic codec deepfakes and SATYAM, a novel hyperbolic ALM that outperforms baselines through dual-stage semantic-prosodic fusion using Bhattacharya distance.
AudioHijack generates imperceptible adversarial audio via gradient estimation, attention supervision, and reverberation blending to hijack 13 LALMs with 79-96% success on unseen contexts and real commercial agents.
RoleJudge is a multidimensional evaluation framework for speech-character alignment in audio LLMs, backed by the RoleChat dataset and multi-stage RL training with standard alignment to reduce reward issues.
HumDial-EIBench is a new benchmark using real human dialogues to evaluate audio language models on emotional intelligence tasks including multi-turn tracking, causal reasoning, empathy generation, and acoustic-semantic conflict resolution.
Ti-Audio is the first multi-dialectal end-to-end Speech-LLM for Tibetan that achieves state-of-the-art performance on ASR and speech translation benchmarks via a Dynamic Q-Former Adapter and cross-dialect cooperation.
Introduces UMUI task for fine-grained multimodal probabilistic inference and CLUE calibration method, where a 3B model matches larger baselines.
Jamendo-MT-QA is a new dataset and benchmark for multi-track comparative music question answering, constructed via an LLM-assisted pipeline from Creative Commons Jamendo tracks and used to evaluate audio-language models.
Speaker-Reasoner is an end-to-end speech LLM that iteratively analyzes audio structure, predicts temporal boundaries, and jointly models speaker identity, gender, timestamps, and transcription using a speaker-aware cache for long audio.
KoALa-Bench is a new public benchmark with six tasks that tests Korean speech recognition, translation, question answering, instruction following, and faithfulness in large audio language models.
Human preferences for the same semantic content show near-chance agreement between text and audio, with audio raters using narrower decision thresholds, less length bias, and more user-oriented criteria.
citing papers explorer
-
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.
-
MusTBENCH: Benchmarking and Advancing Temporal Grounding in Music LLMs
MusTBENCH evaluates temporal grounding in large audio-language models via five expert-validated tasks, and MusT improves performance through encoder adaptation, LLM adaptation, supervised fine-tuning, and RL optimization.
-
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.
-
MIST: Multimodal Interactive Speech-based Tool-calling Conversational Assistants for Smart Homes
MIST is a new synthetic speech-based tool-calling dataset for IoT devices that exposes performance gaps between open- and closed-weight multimodal LLMs.
-
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.
-
KoALa-Bench: Evaluating Large Audio Language Models on Korean Speech Understanding and Faithfulness
KoALa-Bench is a new public benchmark with six tasks that tests Korean speech recognition, translation, question answering, instruction following, and faithfulness in large audio language models.
-
MCGA: A Multi-task Classical Chinese Literary Genre Audio Corpus
MCGA is a new 119-hour multi-task audio corpus for classical Chinese literary genres that shows current MLLMs face substantial challenges on its test set.
-
Hearing to Translate: The Effectiveness of Speech Modality Integration into LLMs
Cascaded systems remain the most reliable for speech translation overall, but recent SpeechLLMs match or outperform them in many conditions while standalone speech models lag.
-
NUTSHELL: A Dataset for Abstract Generation from Scientific Talks
NUTSHELL is a new open dataset of ACL talks paired with abstracts, accompanied by baselines that demonstrate training benefits for speech-to-abstract generation while highlighting remaining challenges.
-
VoiceBench: Benchmarking LLM-Based Voice Assistants
VoiceBench is the first benchmark for multi-faceted evaluation of LLM voice assistants using real and synthetic spoken instructions with speaker, environmental, and content variations.
-
ProactiveLLM: Learning Active Interaction for Streaming Large Language Models
ProactiveLLM enables active interaction in streaming LLMs by learning semantic sufficiency cues from partial inputs through mask-based modeling and synchronized privileged self-distillation without external supervision.
-
Learning When to Think While Listening in Large Audio-Language Models
A wait-think-answer controller for LALMs is trained via SFT followed by six-reward DAPO, raising row-weighted accuracy from 67.6% to 70.3% and cutting post-endpoint thinking length by 14% on synthetic spoken QA while remaining functional on real recorded audio.
-
Locality Matters for Training-Free Audio Token Compression in Audio-Language Models
LTBM merges similar audio tokens under a temporal locality constraint and shows task-dependent advantages over global merging for captioning versus multiple-choice understanding tasks in experiments with Qwen2-Audio and Audio Flamingo 3.
-
Mind the Pause: Disfluency-Aware Objective Tuning for Multilingual Speech Correction with LLMs
A sequence-tagger-guided LLM with contrastive objective corrects disfluencies in Hindi, Bengali, and Marathi ASR transcripts, outperforming removal-only baselines.
-
Hard to Be Heard: Phoneme-Level ASR Analysis of Phonologically Complex, Low-Resource Endangered Languages
Phoneme-level analysis of ASR on Archi and Rutul shows data scarcity explains recognition errors better than phonological complexity, with language-specific adaptations improving wav2vec2 performance.
-
ASPIRin: Action Space Projection for Interactivity-Optimized Reinforcement Learning in Full-Duplex Speech Language Models
ASPIRin decouples speaking timing from token content via binary action space projection and applies GRPO with rule-based rewards to optimize interactivity in SLMs without semantic collapse or repetition.
-
MCAT: Scaling Many-to-Many Speech-to-Text Translation with MLLMs to 70 Languages
MCAT scales MLLMs to many-to-many speech translation across 70 languages via curriculum learning and a 30-token speech adapter, surpassing prior SOTA on FLEURS while improving speed.
-
A cross-species neural foundation model for end-to-end speech decoding
A cross-species pretrained neural encoder combined with end-to-end training and audio LLMs reduces word error rate in neural speech decoding from 24.69% to 10.22% while aligning attempted and imagined speech.
-
Benchmarking Gaslighting Attacks Against Speech Large Language Models
Gaslighting attacks using Anger, Cognitive Disruption, Sarcasm, Implicit, and Professional Negation strategies cause a 24.3% average accuracy drop in Speech LLMs while also triggering behavioral changes like apologies and refusals.
-
Qwen3-Omni Technical Report
Qwen3-Omni is a unified multimodal model that achieves open-source SOTA on 32 of 36 audio and audio-visual benchmarks and overall SOTA on 22 without degrading performance on text, image, or video relative to single-modal Qwen counterparts.
-
Training-Free Multimodal Large Language Model Orchestration
LLM Orchestration integrates modality experts via an LLM controller, cross-modal memory, and interaction layer to enable multimodal input-output without gradient-based training.
-
Step-Audio 2 Technical Report
Step-Audio 2 integrates a latent audio encoder, reasoning-centric reinforcement learning, and discrete audio token generation into language modeling to deliver state-of-the-art performance on audio understanding and conversational benchmarks.
-
StressTest: Can YOUR Speech LM Handle the Stress?
Speech language models fail at reasoning about sentence stress but improve after fine-tuning on a new 17k-example synthetic dataset that varies stress to alter meaning.
-
Step-Audio: Unified Understanding and Generation in Intelligent Speech Interaction
Step-Audio introduces a 130B-parameter unified speech-text model with open-sourced components for understanding, generation, affordable voice cloning, and dynamic control, claiming SOTA human evaluation results on a new benchmark.
-
SALSA: Speech Aware LLM Adaptation via Learned Steering Activation Vectors
SALSA adapts speech-aware LLMs via supervised layer-wise steering vectors, reporting up to 46.8% relative gains over zero-shot on out-of-domain speech benchmarks.
-
Thinking-while-speaking: A Controlled, Interleaved Reasoning Method for Real-Time Speech Generation
InterRS enables real-time speech generation with interleaved reasoning via a controlled data pipeline, interleaved SFT, and RL using TA-Balance and Linguistic Quality rewards, yielding 13% gains on math and logic benchmarks.
-
Minimizing Modality Gap from the Input Side: Your Speech LLM Can Be a Prosody-Aware Text LLM
TextPro-SLM reduces the speech-text modality gap by feeding an LLM backbone with synchronized text tokens and prosody embeddings from WhisperPro, achieving lowest gap scores at 3B/7B scales with roughly 1,000 hours of audio.
-
AUDITA: A New Dataset to Audit Humans vs. AI Skill at Audio QA
AUDITA is a challenging audio QA benchmark where humans score 32% accuracy on average while state-of-the-art models score below 9%, using IRT to reveal systematic model deficiencies.
-
Detecting Hallucinations in SpeechLLMs at Inference Time Using Attention Maps
Four attention metrics enable logistic regression classifiers that detect hallucinations in SpeechLLMs with up to +0.23 PR-AUC gains over baselines on ASR and translation tasks.
-
FreezeEmpath: Efficient Training for Empathetic Spoken Chatbots with Frozen LLMs
FreezeEmpath achieves emotionally expressive speech output and strong performance on empathetic dialogue, speech emotion recognition, and spoken QA tasks by training with a frozen LLM on existing speech datasets.
-
Qwen3.5-Omni Technical Report
Qwen3.5-Omni scales an omnimodal model to hundreds of billions of parameters with 256k context, introduces ARIA for stable speech synthesis, and reports SOTA performance on 215 audio-visual benchmarks while adding multilingual and audio-visual coding capabilities.
-
Do We Need Distinct Representations for Every Speech Token? Unveiling and Exploiting Redundancy in Large Speech Language Models
Deep layers of speech language models show high token redundancy that can be compressed via training-free similarity pooling, reducing prefilling costs by 27% while preserving task performance.
-
A Simple Method to Enhance Pre-trained Language Models with Speech Tokens for Classification
Lasso-selected speech tokens enhance text LLMs for multimodal classification by reducing long audio sequences to task-relevant features via self-supervised adaptation.
-
Enhancing Speech Large Language Models through Reinforced Behavior Alignment
Reinforced Behavior Alignment (RBA) uses self-synthesized data from a teacher LLM and reinforcement learning to close the instruction-following gap in SpeechLMs, outperforming distillation and reaching SOTA on spoken QA and speech-to-text translation benchmarks.
-
Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs
Phi-4-Mini achieves strong math and coding performance with only 3.8B parameters via high-quality synthetic data, while Phi-4-Multimodal uses Mixture-of-LoRAs to integrate modalities and top speech recognition leaderboards.
-
Robust Audio-Text Retrieval via Cross-Modal Attention and Hybrid Loss
A cross-modal attention refinement module plus hybrid loss improves robustness of audio-text retrieval on noisy and long-form audio.
-
On The Landscape of Spoken Language Models: A Comprehensive Survey
A literature survey that organizes spoken language models by architecture, training, and evaluation choices and identifies key challenges and future directions.