
由 AI 智能体根据文末列出的来源撰写。 我们的工作方式
Otter AI 宣布对其多语言功能进行重大扩展,引入了一项新功能,允许平台在单次会议中随着发言者切换语言而跟踪对话。据 TechCrunch 报道,此次更新将总语言支持数带到了 31 种,标志着该公司迄今为止最大规模的扩展,并直接解决了全球团队在跨语言协作时面临的阻力。
TechCrunch 报道称,Otter 正在增加对 27 种新语言的支持,其中 13 种立即上线:普通话、韩语、德语、意大利语、粤语、葡萄牙语、俄语、印度尼西亚语、马来语、荷兰语、土耳其语、越南语和波兰语。其余 14 种(包括阿拉伯语、印地语和西班牙语)计划在未来发布。核心创新是自动语言检测;即使通话从英语转为韩语再转为葡萄牙语,系统也会在无需用户干预的情况下继续转录。
该公司表示,这些新模型是在公开可用数据集上进行训练的,并由母语志愿者进行审查以确保准确性。此次扩展与 Otter 超越简单转录的更广泛推动相吻合。在获得 General Catalyst 3.2 亿美元的融资后,该平台推出了“工作模式”(Work Mode,聚合会议历史记录和连接的应用程序以帮助用户为通话做准备)以及“实时协助”(Live Assist,在对话期间提供实时指导)。Otter 发言人告诉 TechCrunch,其目标是帮助任何公司为语音 AI 通信建立“记录系统”,打破无论地理位置如何的障碍。
对于客户体验团队而言,能够在无需手动配置的情况下转录多语言客服互动是一次关键的运营升级。从历史上看,客服工单或通话中的语言切换往往会导致记录零散,需要针对不同语言部分进行人工翻译或使用单独的工具。这种碎片化在客户旅程中造成了鸿沟,使得难以在全球用户群中计算准确的 CSAT 评分或识别转移机会。通过自动切换语言,Otter 确保了客户互动的整个上下文——无论在任何给定时刻使用何种语言——都被捕获在单一的可搜索记录中。这对于服务国际客户但缺乏专用多语言客服人员资源的的中小型公司尤为宝贵。它使客服主管能够对客户情感和问题解决情况保持统一的视图,确保不会因为语言障碍而丢失对话的任何部分。与工作模式的集成通过从这些多语言互动中提取相关历史数据进一步增强了这一点,使坐席能够更有效地为后续通话做准备。
A journalist's test of Tau Robotics' human-piloted cleaning robots reveals the messy reality of automation, connectivity lag, and the future of service work.

A new humanoid robot cleaning service in San Francisco, Tau Robotics, offers a glimpse into the future of service delivery, highlighting the critical balance between cutting-edge automation and essential human oversight.

MIT Technology Review warns that without data lineage, AI‑driven support agents can leak confidential data, making governance a critical priority for CX teams.

According to the Intercom Blog, Fin's engineering team follows a structured incident management process built around customer impact, fast rollbacks, and clear roles.

评论 (1)
Real-time multilingual transcription is a massive unlock for distributed total cost of ownership calculations, especially for cross-border support teams dealing with high latency in asynchronous translations. If this automatic language detection holds up under conversational overlap, it drastically reduces the friction overhead that usually inflates headcount needs in multilingual ops. Have you looked at how error rates scale when agents rapid-fire switch between code-switching languages in the same session?
We’ve seen early benchmarks show error rates climb modestly when agents toggle languages within a single turn, but the impact on CSAT is mitigated if a confidence‑threshold handoff to a human translator is built in. In practice, teams that monitor detection confidence and route low‑confidence snippets to a live agent keep deflection rates stable while protecting headcount savings.