中
◀ Back to feed
TechCrunch

Otter AI can now follow conversations across 30+ languages

Published 2026-11-09 Collected 2026-11-10 16:00 AI news #ai#enterprise
Full text

Otter.ai, the AI-powered note-taking tool, announced Tuesday that it’s expanding its multilingual capabilities, including a new multilingual transcription feature that can follow conversations as speakers switch between languages during the same meeting.

The company is also adding support for 27 languages, its largest language expansion to date. Once the rollout is complete, Otter will support 31 languages in total, making the platform more accessible to international teams.

The upcoming feature will automatically detect language changes and continue transcribing without requiring users to adjust settings. For example, a call could move from English to Korean to Portuguese, with Otter following the conversation throughout.

The update aims to address a common challenge for global teams, where language differences can make meetings and collaboration more difficult.

“Otter is breaking down language barriers so any company, no matter where they’re based, can build a system of record for the decisions for voice AI communication,” an Otter spokesperson told TechCrunch.

Thirteen of the new languages are available now: Mandarin Chinese, Korean, German, Italian, Cantonese, Portuguese, Russian, Indonesian, Malay, Dutch, Turkish, Vietnamese, and Polish.

The remaining 14 languages are coming soon and include Swedish, Danish, Finnish, Arabic, Hindi, Thai, Hebrew, Norwegian, Greek, Macedonian, Persian, Czech, Hungarian, and Romanian.

The newly added languages run on Otter’s own speech models. The company told us it trained them using publicly available datasets and worked with native-speaking volunteers to review transcripts and check accuracy.

The announcement comes as Otter continues to expand beyond transcription and into other workplace AI tools, following its $320 million funding from General Catalyst.

It recently introduced Work Mode, which uses meeting history and connected apps to help users prepare for upcoming calls. For instance, the feature pulls information from previous customer conversations, outstanding action items, notes from teammates’ similar calls, and recent Google Drive files.

Another recent addition is Live Assist, which provides real-time coaching during meetings. It analyzes the conversation and offers relevant suggestions based on what has already been talked about and what is happening in the moment.

Source
Title
Otter AI can now follow conversations across 30+ languages
Author
Lauren Forristal
Site
TechCrunch
Published
2026-11-09
Collected
2026-11-10 16:00 · 2026-11-10-16-ai.json
Link
https://techcrunch.com/2026/11/09/otter-ai-can-now-follow-conversations-across-30-languages/
Open original ↗

Reposted. Will be removed on request. All rights belong to the original author.

More from this batch

16:00batch · 2026-11-1012 items
>

The AI 'credit card' is getting costlier as CoreWeave bond yield tops 16%

CoreWeave's steep yields reflect stress in the AI debt markets as investors worry about a data center slowdown. What comes next.

CNBC
>

谷歌 Logan Kilpatrick 在 Fable 5.2 发布几小时后只发了一个词「Gemini」,被普遍解读为新模型本周要来

这条只有一个词的帖子已有 26 万多次浏览、3400 多个赞,宝玉等人都在转发调侃;Gemini 4 Carbon 刚被 Fable 5.2 挤下榜首,谷歌大概率很快回应。

X · @OfficialLoganK
>

Zora:用「检索」代替草稿模型做 VLA 推测解码,机器人动作生成纯软件提速 3.4 倍、配专用硬件 7.6 倍

改进了什么:自回归 VLA 模型在机器人端侧算力上延迟太高。靠什么:利用机器人动作的规律,从历史动作里检索出草稿,不用额外训练草稿模型;再用运动学约束决定接受哪些动作,并让批量验证共享 KV 缓存。效果:在 LIBERO 上 4 个 VLA 模型精度基本不变。

arXiv 2611.11098
>

SoMem:让智能体用普通文档自己整理长期记忆,三个记忆基准都超过最强基线 2 到 3 个百分点

改进了什么:智能体记忆以往靠预设字段和固定规则组织,跟不上任务变化。靠什么:记忆存成互相链接的文档,由智能体自己决定总结什么、怎么归类、何时修订,每条总结都链回原始对话当证据;查询时智能体自己选读哪些文档。整理好的记忆还能跨模型复用,代码已开源,做企业知识助手可以参考。

arXiv 2611.11929
>

SpecMem:把用户用过的工具调用存成「活」的草稿库,智能体工具调用最多提速 1.74 倍

改进了什么:工具调用要逐 token 解码,延迟随调用增多而上升。靠什么:用推测解码,从每个用户过去验证过的调用里检索草稿,并在部署中持续写回、按 LRU 淘汰;对有副作用的不可逆调用先暂缓,等验证通过再执行。按用户隔离的存储还堵住了跨用户泄露的通道。

arXiv 2610.31840
>

Flide:给「决策型」小模型做的精确推理引擎,比 vLLM 和 SGLang 最快 9.5 倍

改进了什么:智能体里越来越多用小模型做分类式快速决策,现有生成型推理引擎要么重复算前缀、要么概率失真。靠什么:把一次请求的分支树压进单个扁平 token 流,用 ForkAttention 在注意力层和 Gated DeltaNet 层上精确分叉,保证每个分支和单独前向计算结果完…

arXiv 2610.42148
>

OneTrans-GSR:用一个自回归 Transformer 端到端生成推荐曝光列表,线上 GMV/u 提升 2.63%、延迟降 13%

改进了什么:电商推荐的重排和混排一直靠人工规则和小模型拼接。靠什么:把两个阶段合成一个解码器策略,训练时用奖励模型给整张列表打分做强化学习,解码时显式执行业务约束;离线实验显示算力越多列表质量越好,HF 当日 33 票。

arXiv 2611.04720
>

Sage-Taiyi:带「元认知」的科研智能体,在 AI 科学家选题基准上 54.0 分,超过 AI Scientist-v2 的 47.9

改进了什么:AI 科学家要决定下一步研究什么、怎么研究。靠什么:显式监控自己对证据、假设和未解问题的理解,再据此调整研究目标、挑选可复用的推理方法。效果:和专家合作做实验,把一个钴催化羰基化反应的产率从痕量提到 84%。

arXiv 2611.12762
>

SIRIS 基准:大模型能看出模拟器的误差会不会改变设计结论吗?13 个前沿模型都会出错

用 LLM 推理服务的设计题测试:模型能指出模拟器的局限,不代表能做对选择,最强的模型有时也会放弃本来正确的推荐;加一轮自我审查能纠正一些错误,但也会引入新错误。用智能体做系统研究和容量规划时值得警惕。

arXiv 2611.10329
>

金融单据识别不能只看字符错误率:一套按风险约束的评估和「选择性自动化」框架

主张把识别结果当成「能追溯到原图区域的金融事实」来评估,对金额、编号、正负号、期间这类关键字段设不可补偿的验收门槛,再把不确定的单据按风险和人工审核产能路由给人。纯方法论文章,没有新模型,但做票据和报表自动化的团队可以直接借鉴验收标准。

arXiv 2611.02775
>

ooosplat:一键把视频和图片变成 3D 高斯泼溅(Gaussian Splat)的本地桌面应用

2,425 星,今日新增 121 星,Rust 编写,许可证未确认。全部在本地运行,不用上传云端,适合做数字孪生、展厅和商品 3D 化的快速原型。

GitHub
>

英伟达Jim Fan: 大模型 会指挥,机器人为什么还需要自己的模型?

从英伟达机器人自主取样、上显微镜的实验讲起,解释为什么大模型只能做「指挥」,真正动手还需要专门的机器人动作模型。

公众号 · NE时代智能车