讯飞AI眼镜OTA4升级 端云协同122语种翻译与AI图文纪要落地iFLYTEK AI Glasses OTA4 Upgrade: Device-Cloud Synergy Brings 122-Language Translation and AI Illustrated Minutes


AI free token/AI免费积分

讯飞AI眼镜OTA4升级 端云协同122语种翻译与AI图文纪要落地

本文来源:https://news.iresearch.cn/yx/2026/11/574159.shtml
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

10月22日,2026 科大讯飞 全球1024开发者节主论坛在北京举行。走到第十个年头的1024,今年释放的核心信号不再是单一模型参数的比拼,而是“端云一体”如何让AI真正落到硬件上、落进场景里。作为这一架构在穿戴形态上的直接承载者,讯飞AI眼镜的亮相让“软硬一体”从战略表述变成了可触摸的产品逻辑。11月9日,这款产品迎来 OTA 4大版本升级,这也是“云+端”新解法的一次系统性交付。

翻译能力进化:从“能翻”到“翻得快、翻得准”

讯飞AI眼镜支持122种语言(含方言及口音)实时翻译,覆盖全球200多个国家和地区,构建了涵盖旁听同传、面对面翻译、线上同传、通话翻译和视觉翻译的全场景翻译体系。值得一提的是,OTA4最硬核的升级在翻译性能,9个小语种的翻译能力完成专项升级,大幅优化同传时延与翻译效果,字幕出字更快,对话同步感更强。转写首响、翻译首响、翻译尾响等耗时参数,缩短约80%。

与此同时,发布会上早已公布、用户期待已久的唇动识别降噪在OTA4中正式上线。在面对面翻译中开启唇动识别后,眼镜正对当前说话人,状态栏出现唇形图标即表示唇形降噪生效――展馆、多人讨论、咖啡厅等有人声干扰的场景下,旁边人声被有效抑制,当前说话人的识别效果显著提升。这是讯飞独创的多模态AI降噪能力,也是普通翻译设备难以复制的技术壁垒。

办公闭环进化:从“对话”到“图文并茂的办公资产”

硬件解决“能不能用”,Agent中枢解决“产出什么”。讯飞AI眼镜内置GlassClaw AI助理,支持多步骤复合指令,接入 腾讯 WorkBuddy后填补“线下实景”拼图。11月9日,OTA4升级引入“AI图文纪要”,进入next level:现场随手拍摄,PPT、主讲人、展品、会场画面尽收其中,照片按对话时序自动归档。散会后Loomy或WorkBuddy就能自动串联全部图文,按时间线完整还原整场交流,快速生成图文并茂的纪要。不论是商务洽谈,还是展会参观、课堂记录、讲座研学,全程一目了然。

配合角色分离功能――旁听同传和线上同传中自动区分发言人,眼镜端和 App 端分段展示“说话人1/2/3”,会后导出记录发言人自动标注――多人跨国会议“字幕堆在一起、会后分不清谁说的”的痛点被化解。手机播报译文则让面对面翻译更人性化:对方无需盯着用户手机屏幕,通过手机外放即可接收同传。仅转写模式在四大翻译场景中均可开启,只录原文不翻译,满足纯记录需求。值得一提的还有,拍照速度也整体优化,新增的预览功能让你拍完照就能看到照片缩略图,当场确认拍没拍上,关键画面一张不落。

过去两年,AI眼镜赛道涌入大量“硬件+第三方大模型API”的拼装产品,能听清却听不懂,能看见却不会思考。讯飞的路径不同:星火大模型全栈自研,从云端训练到端侧推理自主可控;AI眼镜作为端侧载体,承担“视野交互”的入口角色;OTA4的翻译性能跃升与多模态纪要落地,则证明“云+端”协同不是概念,而是可以按月迭代的产品节奏。

放在1024十年节点上看,这条路线指向一个清晰判断:大模型的下半场,不在排行榜上,在用户鼻梁上、在会议桌上、在每一次跨语言沟通的现场。OTA4,是这一判断的最新注脚。

上一篇: 嘉盛汇JSH FX坚持长期主义理念,走可持续金融科普路线

OpenAI 开放 Decisions API:只对输入 token 收费,输出不再计费 2026/10/23 14:45:21

Anthropic 把 Agent Skills 标准捐给 Linux 基金会旗下 Agentic AI Foundation 2026/10/23 15:03:59

星巴克 2026 财年 Q4:全球同店增 7.2%,中国业务转特许合资后首份年报 2026/11/6 17:42:56

Cloudflare 与 Commerce.com 让近 10 万个店面接上 AI 智能体 2026/10/22 14:22:47

Mistral 开放 Mistral Large 4 权重:1.05 万亿参数,单任务少烧 47% token,API 价格腰斩 2026/11/6 17:08:09

OpenRouter 周榜(数据截至 2026-10-26):DeepSeek V4.1 Flash 1017 居首 2026/10/27 11:45:09

快速导航 互联网+ 营销 行业库 专题 图文资讯

Copyright© 沪公网安备 31010402000581号 沪ICP备15021772号-10

2002 - Copyright© 上海艾瑞市场咨询股份有限公司 沪公网安备

iFLYTEK AI Glasses OTA4 Upgrade: Device-Cloud Synergy Brings 122-Language Translation and AI Illustrated Minutes

本文来源:https://news.iresearch.cn/yx/2026/11/574159.shtml
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

On October 22, the main forum of the 2026 iFLYTEK Global 1024 Developer Festival was held in Beijing. In its tenth year, the core signal sent by this year's 1024 was no longer a contest over individual model parameters, but how “device-cloud integration” can truly bring AI onto hardware and into real-world scenarios. As the direct wearable embodiment of this architecture, the debut of iFLYTEK AI Glasses turned “software-hardware integration” from a strategic statement into a tangible product logic. On November 9, the product received a major OTA 4 upgrade—a systematic delivery of the new “cloud + device” solution.

Evolution of translation capabilities: from “being able to translate” to “translating fast and accurately”

iFLYTEK AI Glasses support real-time translation in 122 languages (including dialects and accents), covering more than 200 countries and regions worldwide, and have built a full-scenario translation system encompassing ambient simultaneous interpretation, face-to-face translation, online simultaneous interpretation, call translation, and visual translation. Notably, the most substantial upgrade in OTA4 is in translation performance: translation capabilities for nine minor languages received targeted upgrades, significantly reducing simultaneous interpretation latency and improving translation quality, with subtitles appearing faster and a stronger sense of conversational synchronization. Timing metrics such as first-token latency for transcription, first-token latency for translation, and translation completion latency have been shortened by approximately 80%.

At the same time, lip-motion recognition noise cancellation—announced at the launch event long ago and eagerly awaited by users—has officially gone live in OTA4. When lip-motion recognition is enabled during face-to-face translation, the glasses face the current speaker; a lip-shaped icon appearing in the status bar indicates that lip-shape noise cancellation is active. In environments with competing voices such as exhibition halls, group discussions, and cafés, nearby voices are effectively suppressed and recognition of the current speaker is significantly improved. This is iFLYTEK's unique multimodal AI noise-cancellation capability, and a technical barrier that ordinary translation devices find hard to replicate.

Evolution of the office workflow: from “conversation” to “illustrated office assets”

Hardware determines “whether it can be used,” while the Agent hub determines “what gets produced.” iFLYTEK AI Glasses come with a built-in GlassClaw AI assistant that supports multi-step compound commands; after integrating Tencent WorkBuddy(了解更多), they complete the “offline real-world scenario” piece of the puzzle. On November 9, the OTA4 upgrade introduced “AI Illustrated Minutes,” taking things to the next level: snap photos on the spot, capturing slides, the speaker, exhibits, and venue scenes, with photos automatically archived in the order of the conversation. After the meeting, Loomy or WorkBuddy can automatically link all the images and text, fully reconstructing the entire exchange along the timeline and quickly generating richly illustrated minutes. Whether for business negotiations, trade show visits, class notes, or lecture and study sessions, everything is clear at a glance.

Combined with the speaker separation feature—which automatically distinguishes speakers during ambient and online simultaneous interpretation, displays “Speaker 1/2/3” in segments on both the glasses and the app, and automatically labels speakers in records exported after the meeting—the pain point of multilingual meetings, where “subtitles pile up and you can't tell who said what afterward,” is resolved. Phone playback of translations makes face-to-face translation more human: the other party doesn't need to stare at the user's phone screen; they can receive the interpretation through the phone's speaker. Transcription-only mode can be enabled in all four major translation scenarios, recording only the original audio without translation to meet pure documentation needs. Also worth noting, photo capture speed has been optimized overall, and a new preview feature lets you see a thumbnail right after taking a photo, confirming on the spot whether the shot succeeded so no key moment is missed.

Over the past two years, the AI glasses sector has seen an influx of assembled products combining “hardware + third-party large-model APIs”—able to hear clearly but not understand, able to see but not think. iFLYTEK takes a different path: the Spark large model is developed in-house across the full stack, from cloud training to on-device inference, all under its own control; the AI glasses serve as the device-side carrier, acting as the entry point for “vision-based interaction”; and OTA4's leap in translation performance and the realization of multimodal minutes prove that “cloud + device” synergy is not a concept but a product cadence that can be iterated monthly.

Viewed against the tenth anniversary of the 1024 Developer Festival, this path points to a clear judgment: the second half of the large-model race is not on leaderboards, but on the bridges of users' noses, on conference tables, and in every real-world cross-language exchange. OTA4 is the latest footnote to this judgment.

Previous article: JSH FX upholds a long-term philosophy, taking a sustainable path in financial literacy education

OpenAI opens up Decisions API: charging only for input tokens, with output no longer billed 2026/10/23 14:45:21

Anthropic donates the Agent Skills standard to the Agentic AI Foundation under the Linux Foundation 2026/10/23 15:03:59

Starbucks FY2026 Q4: global same-store sales up 7.2%, first annual report after the China business shifted to a franchised joint venture 2026/11/6 17:42:56

Cloudflare and Commerce.com connect nearly 100,000 storefronts to AI agents 2026/10/22 14:22:47

Mistral releases Mistral Large 4 weights: 1.05 trillion parameters, 47% fewer tokens burned per task, API price slashed in half 2026/11/6 17:08:09

OpenRouter weekly rankings (data as of 2026-10-26): DeepSeek V4.1 Flash 1017 takes the top spot 2026/10/27 11:45:09

Quick navigation: Internet+ Marketing Industry Library Topics Illustrated News

Copyright© Shanghai Public Network Security Record No. 31010402000581 Shanghai ICP Record No. 15021772-10

2002 - Copyright© Shanghai iResearch Market Consulting Co., Ltd. Shanghai Public Network Security Record

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry