Primary sourcerohanpaul_ai
View sourceEvent ·Multi-source
VoiceArena 推出 Diarization Bench,同规则评测 12 款说话人分离系统
VoiceArena 发布 Diarization Bench,用同一批真实音频和统一打分规则对比 12 款 diarization(说话人分离)系统。基准显示,重叠语音和说话人数量增加时,各系统准确率下降明显。评测还揭示 collar(时间容差)对得分的影响:NVIDIA Nemotron 3 在 250 ms collar 下 DER 为 4.29%,同一输出按 0 ms collar 评分则升至 14.72%。该基准补充了 VoiceArena 已有的 STT、TTS 和对话智能体评测。
Current read
VoiceArena 搞了个 diarization 评测榜,12 款系统同题同规则比拼,还能看时间容差一变分数差多大,做语音的可以去看看。
11 sources58° AI heatLast updated 2026/9/23 17:06:36
Evidence
11 sourcesRelated source 1NVIDIA AI
View sourceRelated source 2Hugging Face: Blog
View sourceRelated source 3AIGCLINK
View sourceRelated source 4Thomas Wolf
View sourceRelated source 5shao__meng
View sourceRelated source 6The Business Times: Tech
View sourceRelated source 7IT之家
View sourceRelated source 8LangChain
View sourceRelated source 9Runway ML
View sourceRelated source 10DAIR.AI
View sourceConflict 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.