Event ·Multi-source

Artificial Analysis 发布 Diarization Index 说话人分离基准

Artificial Analysis 推出 Diarization Index,用于衡量语音转文字 API 的说话人分离能力,核心指标为 Word Diarization Error Rate(WDER),即被分配给错误说话人或未分配说话人的词占比。 基准覆盖 97 段对话,人数从 2 人到 8 人不等,并对每种人数规模赋予相同权重,避免偏向单一场景。 首轮评测中,NVIDIA 的 Nemotron 3.5 Transcribe 以 3.3% 的 WDER 排名第一,Mistral 的 Voxtral Small Transcribe 3 以 7.5% 位居第二,两者差距超过一倍。

Current read

做会议纪要、访谈转写的朋友可以看看这个新基准,专门测“哪句话是谁说的”。NVIDIA 的 Nemotron 3.5 Transcribe 错误率只有 3.3%,比 Voxtral Small Transcribe 3 的 7.5% 低一半多,选 API 前值得对照。

11 sources43° AI heatLast updated 2026/11/13 21:17:49

Evidence

11 sources
Primary sourceArtificial Analysis
View source
Related source 2Sebastian Raschka
View source
Related source 3CoreWeave
View source
Related source 4AWS Machine Learning Blog
View source
Related source 5Amazon Science
View source
Related source 6TestingCatalog
View source
Related source 7Mira Murati (TML)
View source
Related source 9LMSYS Org (SGLang)
View source
Related source 10Jerry Liu
View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.