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.
hub Canonical reference
Step-Audio 2 Technical Report
Canonical reference. 82% of citing Pith papers cite this work as background.
50
Pith papers citing it
Background
82% of classified citations
abstract
This paper presents Step-Audio 2, an end-to-end multi-modal large language model designed for industry-strength audio understanding and speech conversation. By integrating a latent audio encoder and reasoning-centric reinforcement learning (RL), Step-Audio 2 achieves promising performance in automatic speech recognition (ASR) and audio understanding. To facilitate genuine end-to-end speech conversation, Step-Audio 2 incorporates the generation of discrete audio tokens into language modeling, significantly enhancing its responsiveness to paralinguistic information such as speaking styles and emotions. To effectively leverage the rich textual and acoustic knowledge in real-world data, Step-Audio 2 integrates retrieval-augmented generation (RAG) and is able to call external tools such as web search to mitigate hallucination and audio search to switch timbres. Trained on millions of hours of speech and audio data, Step-Audio 2 delivers intelligence and expressiveness across diverse conversational scenarios. Evaluation results demonstrate that Step-Audio 2 achieves state-of-the-art performance on various audio understanding and conversational benchmarks compared to other open-source and commercial solutions. Please visit https://github.com/stepfun-ai/Step-Audio2 for more information.
hub tools
citation-role summary
background 9
baseline 1
method 1
citation-polarity summary
representative citing papers
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.
RedVox benchmark shows speech model safety and fairness vulnerabilities persist under non-adversarial conditions, worsen in non-English languages, and increase with spoken inputs.
Introduces the first benchmark for over-refusal in large audio language models using 3,000 pseudo-harmful audio samples and evaluates 12 models across six families, finding widespread over-refusal.
Instruction-based vector steering redirects temporal attention in LALMs to acoustically relevant regions, recovering queried sound event locations with 60.87-68.72% overlap accuracy without training.
FD-SLMs exhibit state inertia during abrupt interruptions that a training-free perception-vector steering intervention mitigates, lifting correctness from 28% to 45% and IWOR from 40% to 72% on the Zero-Buffer Benchmark.
SpeechEditBench provides seven atomic editing tasks, compositional multi-operation instructions, and an anchor-based protocol yielding target success, preservation success, and joint success metrics; evaluations show no model excels across dimensions and compositional editing is especially difficult
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.
DuplexSLA introduces a three-channel full-duplex architecture that synchronizes continuous user audio, discrete assistant audio, and rate-limited textual actions inside a single backbone for native turn-taking and in-conversation tool use.
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.
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.
LWS is a text-first paradigm for full-duplex speech LLMs that treats visible writing as a primary output channel alongside audio input and spoken response, implemented via token schema and synthetic per-second annotations.
SpeechParaling-Bench is a new evaluation framework for paralinguistic-aware speech generation that reveals major limitations in current large audio-language models.
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.
CapTalk unifies single-utterance and dialogue voice design via utterance- and speaker-level captions plus a hierarchical variational module for stable timbre with adaptive expression.
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.
TiCo enables spoken dialogue models to follow explicit time constraints in generated responses using Spoken Time Markers and reinforcement learning with verifiable rewards, cutting duration error by 2.7x over its backbone.
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.
Spoken language models exhibit style amnesia and fail to maintain instructed paralinguistic styles across multi-turn conversations, with explicit recall offering partial mitigation.
Open-ended evaluation with real speech reveals that gender and accent cues cause statistically significant, task-dependent distributional shifts in recommendations from 12 large audio-language models.
Lychee-FD resolves modality interference in full-duplex spoken language models by separating acoustic and semantic parameters in deep layers and adding a dense semantic alignment channel, achieving state-of-the-art performance on spoken QA and interaction benchmarks.
Audex unifies audio understanding and generation on a strong text MoE backbone with multi-stage SFT plus text-only Cascade RL, matching open SOTA audio scores while mostly retaining text capability.
PRIME-Speech adds low-latency speech output to frozen S2T LLMs by synchronizing a causal post-decoder with intermediate hidden states and using mixed conditioning plus turn-level KV-cache packing, preserving original S2T performance across translation, QA, and dialogue tasks.
MSU-Bench is a new two-tier benchmark covering speaker grounding to dialogue reasoning in multi-speaker conversations, with Gemini-assisted annotation and human verification.
citing papers explorer
-
Rethinking Entropy Allocation in LLM-based ASR: Understanding the Dynamics between Speech Encoders and LLMs
A multi-stage training method for LLM-based ASR uses new entropy allocation metrics to achieve competitive benchmark performance with 2.3B parameters while mitigating hallucinations via better encoder-LLM decoupling.