Event ·Multi-source

ElevenLabs 发布 Scribe v3 与 Scribe v3 Live 语音识别模型

ElevenLabs 发布了 Scribe v3 和 Scribe v3 Live 两款语音识别模型,主打在背景噪音和多人同时说话的场景下保持准确转写。根据 Artificial Analysis 的基准测试,Scribe v3 的准确率小幅超过了 OpenAI、Google 和 Microsoft 的同类模型。 Scribe v3 Live 针对实时语音智能体场景做了专门优化,它会利用对话上下文来判断说话者是否已经说完,避免智能体在来电者还在讲话时就打断对方,这是电话客服类应用中常见的痛点。 不过这套精度是有代价的:Microsoft 和 xAI 的模型运行速度快数倍,价格大约只有一半。对延迟敏感、对成本敏感的团队需要在准确率与速度、价格之间权衡。

Current read

做电话语音智能体的可以看看,ElevenLabs 新版 Scribe v3 Live 会结合上下文判断用户是否说完,能减少打断式抢话。准确率在 Artificial Analysis 测试里略胜 OpenAI 和 Google,但速度和价格不如 Microsoft 和 xAI,需要按场景取舍。

11 sources51° AI heatLast updated 2026/11/12 13:48:19

Evidence

11 sources
Related source 1Artificial Analysis
View source
Related source 2ElevenLabs
View source
Related source 3Gorden Sun
View source
Related source 4Google Gemini App
View source
Related source 5Mustafa Suleyman
View source
Related source 6Shubham Saboo
View source
Related source 7Databricks
View source
Related source 8Rowan Cheung
View source
Related source 9TestingCatalog
View source
Related source 10theverge
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.