VoxSafeBench reveals that speech language models recognize social norms from text but fail to apply them when acoustic cues like speaker or scene determine the appropriate response.
hub Mixed citations
CosyVoice 3: Towards In-the-wild Speech Generation via Scaling-up and Post-training
Mixed citation behavior. Most common role is background (33%).
29
Pith papers citing it
Background
33% of classified citations
abstract
In our prior works, we introduced a scalable streaming speech synthesis model, CosyVoice 2, which integrates a large language model (LLM) and a chunk-aware flow matching (FM) model, and achieves low-latency bi-streaming speech synthesis and human-parity quality. Despite these advancements, CosyVoice 2 exhibits limitations in language coverage, domain diversity, data volume, text formats, and post-training techniques. In this paper, we present CosyVoice 3, an improved model designed for zero-shot multilingual speech synthesis in the wild, surpassing its predecessor in content consistency, speaker similarity, and prosody naturalness. Key features of CosyVoice 3 include: 1) A novel speech tokenizer to improve prosody naturalness, developed via supervised multi-task training, including automatic speech recognition, speech emotion recognition, language identification, audio event detection, and speaker analysis. 2) A new differentiable reward model for post-training applicable not only to CosyVoice 3 but also to other LLM-based speech synthesis models. 3) Dataset Size Scaling: Training data is expanded from ten thousand hours to one million hours, encompassing 9 languages and 18 Chinese dialects across various domains and text formats. 4) Model Size Scaling: Model parameters are increased from 0.5 billion to 1.5 billion, resulting in enhanced performance on our multilingual benchmark due to the larger model capacity. These advancements contribute significantly to the progress of speech synthesis in the wild. We encourage readers to listen to the demo at https://funaudiollm.github.io/cosyvoice3.
hub tools
citation-role summary
background 2
method 2
baseline 1
dataset 1
citation-polarity summary
representative citing papers
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.
PlanAudio introduces a unified autoregressive LLM framework with semantic latent chain-of-thought for generating composite speech and sound audio from free-form text, plus a new benchmark.
GibbsTTS combines a training-free kinetic-optimal scheduler with finite-step moment correction in MI-DFM to deliver top naturalness and strong speaker similarity in zero-shot TTS.
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.
A new dataset, iterative coarse-to-fine localization framework, and segment-level IoU F1 metric tackle the open problem of detecting multiple unknown word-level inpainted regions in speech.
Semantic priors from HuBERT and Whisper improve speech codec intelligibility up to 6 kbps but show diminishing returns beyond that, with a bitrate-aware regulation strategy balancing semantic consistency and naturalness.
AST enables seamless speech editing by latent recomposition on pre-trained TTS models plus adaptive weak fact guidance, plus a new dataset and WDTW metric, claiming 70% WER reduction and better temporal consistency without training.
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.
CAST benchmark shows language models infer correct word stress from discourse context but TTS systems frequently fail to produce it in speech.
SQuTR aggregates 37k queries from six text retrieval datasets, synthesizes speech from 200 speakers, adds 17 noise categories at varying SNR, and shows that even large retrieval models degrade sharply under extreme acoustic noise.
ViBES introduces a speech-language-behavior model using modality-specific transformer experts that jointly generates dialogue and 3D body actions, showing gains over separate co-speech and text-to-motion baselines on multi-turn metrics.
LaSR improves context-aware terminology recognition in speech LLMs by aligning latent CoT supervision on acoustic regions and introducing latent reasoning periods, shown on a new academic corpus to outperform standard fine-tuning without added latency.
RobustSpeechFlow improves TTS alignment robustness by extending contrastive flow matching with length-preserving repeat and skip latent augmentations, lowering WER from 1.44 to 1.38 on Seed-TTS-eval and CER on ZERO500.
SemaVoice adds SFM-guided alignment to refine continuous speech representations in autoregressive TTS, reporting 1.71% English WER on Seed-TTS and competitiveness with open-source SOTA.
S2ST-Omni 2 uses typology-informed hierarchical encoding, gated Dual-CTC, and typology-aware prompting to improve multilingual S2ST over flat-label baselines on CVSS-C, with gains in low-data regimes.
TTS-PRISM defines a 12-dimensional perceptual schema, builds a targeted diagnostic dataset via adversarial synthesis and expert labels, and tunes an end-to-end model that outperforms generalist LLMs in human alignment on a 1,600-sample Mandarin test set while profiling six TTS paradigms.
UAF is the first unified audio front-end LLM that turns multiple front-end tasks into one sequence prediction model processing streaming audio chunks and reference prompts to output semantic and control tokens for full-duplex interaction.
MoVE uses specialized LoRA expert adapters and a soft router to translate non-verbal vocalizations in S2ST, reproducing them in 76% of cases versus at most 14% for baselines while scoring highest on naturalness and emotional fidelity.
Audio2Tool is a new benchmark dataset that shows speech models perform well on simple commands but degrade sharply on compositional tasks and realistic acoustic noise.
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.
OmniVoice introduces a diffusion language model-style non-autoregressive TTS system that directly maps text to multi-codebook acoustic tokens, scaling zero-shot synthesis to over 600 languages with SOTA results on multilingual benchmarks using 581k hours of open data.
FoleyDirector introduces structured temporal scripts and a fusion module to enable precise timing control in DiT-based video-to-audio generation while preserving audio fidelity.
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.