Research map / Image generation / Autoregressive and latent generation
VibeVoice: Expressive Podcast Generation with Next-Token Diffusion
Zhiliang Peng, Jianwei Yu, Wenhui Wang, Yaoyao Chang, Yutao Sun, Li Dong, Yi Zhu, Weijiang Xu, Hangbo Bao, Zehua Wang, Shaohan Huang, Yan Xia and 1 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add semantic tokenizer to acoustic tokenizer (hybrid) (ablation) | VibeVoice-1.5B · VibeVoice-Eval short, overall | WER-W: 6.22 -> 1.84 | Improved |
| add semantic tokenizer to acoustic tokenizer (hybrid) (ablation) | VibeVoice-1.5B · VibeVoice-Eval short, overall | SIM-O: 0.68 -> 0.64 | Hurt |
| use hybrid tokenizer instead of coupled tokenizer (ablation) | VibeVoice-1.5B · VibeVoice-Eval short, overall | SIM-O: 0.45 -> 0.64 | Improved |
| use hybrid tokenizer instead of coupled tokenizer (ablation) | VibeVoice-1.5B · VibeVoice-Eval short, overall | WER-W: 3.55 -> 1.84 | Improved |
| scale model from 1.5B to 7B (ablation) | VibeVoice · VibeVoice-Eval short, overall | WER-W: 2.11 -> 0.66 | Improved |
| scale model from 1.5B to 7B (ablation) | VibeVoice · VibeVoice-Eval short, overall | SIM-O: 0.59 -> 0.75 | Improved |
| scale model from 1.5B to 7B (ablation) | VibeVoice · Podcast subjective | average MOS: 3.54 -> 3.76 | Improved |
| increase LLM training sequence length from 32K to 64K (ablation) | VibeVoice-1.5B · VibeVoice-Eval short, overall | WER-W: 2.11 -> 1.22 | Improved |
| increase LLM training sequence length from 32K to 64K (ablation) | VibeVoice-1.5B · VibeVoice-Eval long, overall | WER-W: 1.33 -> 1.55 | Hurt |
| use 10 DDPM steps instead of 5 with CFG 1.25 (ablation) | VibeVoice · VibeVoice-Eval | WER | Improved |
| increase DDPM steps beyond 5 (ablation) | VibeVoice · VibeVoice-Eval | SIM-O | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MoonCast — MoonCast: High-Quality Zero-Shot Podcast Generation (NeurIPS 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MoonCast: High-Quality Zero-Shot Podcast Generation (NeurIPS 2025 · similarity 0.90)
- CoVoMix2: Advancing Zero-Shot Dialogue Generation with Fully Non-Autoregressive Flow Matching (NeurIPS 2025 · similarity 0.88)
- Generative Audio Language Modeling with Continuous-valued Tokens and Masked Next-Token Prediction (ICML 2025 · similarity 0.88)
- CoVoMix: Advancing Zero-Shot Speech Generation for Human-like Multi-talker Conversations (NeurIPS 2024 · similarity 0.87)
- NaturalSpeech 2: Latent Diffusion Models are Natural and Zero-Shot Speech and Singing Synthesizers (ICLR 2024 · similarity 0.86)
- AudioCALM: Continuous Autoregressive Language Modeling for Universal Audio Generation (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
