Underdog 新演示展示 32MB 语音模型

全球的 AI 公司都在竞相打造更大、更好的模型,但问题在于,它们全都需要海量的算力。而在另一端,一些研究人员和初创公司正在打造更小的模型,让它们能在手机或电脑等边缘设备上运行。
其中一家名为 Underdog 的初创公司发布了一款游戏,它是用该公司的新语音模型 Woof 打造的,而该模型只有 32MB 大小。
这款游戏呈现了一个复古科技主题的房间,里面有一台打字机和一个 CRT 显示器。还有一盏台灯,点击它的开关会发出声音。开场提示会警告用户,镇上存在“侵入性”的云服务提供商——表面上看是在暗讽基于云端的模型。
你可以与打字机和音乐调谐器互动。
我们为我们的新前沿文本转语音模型 Woof 做了一款游戏。31.7 MB,小到可以在你的浏览器里运行 在 X 上玩这款游戏……所有声音都是 Woof。击败 Darkcloud 即可获得邀请 https://t.co/ImUFYDCBQGpic.twitter.com/tthHx7ZGBK
— Underdog AI (@UnderdogAI) 2026年11月5日
你可以用键盘在打字机上输入任何内容,游戏会把它们念出来。你还可以旋转调谐器来生成不同种类的音乐。
所有这些都在浏览器内通过 Underdog 的语音生成模型运行。该公司称,该模型可以在大多数设备上运行,且无需任何网络连接。该模型有 3100 万个参数,并开放给所有人下载。
尽管这么小的模型可能无法取代更大的模型,但它或许仍足以支撑本地助手体验,或用于生成个性化播客。
这一演示发布数周前,Underdog 刚推出 Bark——一个 167 MB 的语音转文本模型,同样设计用于在边缘设备上运行,以完成转写和听写。
Underdog 创始人 Sigil Wen 对 TechCrunch 表示,他之所以开始开发这款应用,是因为他不想使用 Granola 这类笔记工具,而是想用可在用户设备上运行的 AI 模型搭建自己的系统,让用户拥有自己的数据。
“我认为,几个月内我们就能在大家现有的设备上运行当今大部分能力最强的模型。我们只想训练最好的开放模型——美国制造、开放、能力强,让人们拥有私密的 AI。”他在谈到自己的宣言时说道。
除了上述两个语音模型,Underdog 还发布了一个可在笔记本电脑上运行的 270 亿参数模型,以及一个需排队等候的应用,公司将其定位为 Muse 和 Instinct 的本地托管替代方案。
当您通过我们文章中的链接购买时,我们可能会获得一小笔佣金。这不会影响我们的编辑独立性。
Ivan 在 TechCrunch 报道全球消费科技动态。他常驻印度,此前曾在 Huffington Post 和 The Next Web 等媒体工作。
您可以通过发送邮件至 [email protected] 联系或核实来自 Ivan 的联系,也可通过 Signal 上的加密消息账号 ivan.42 与他联系。
正在加载下一篇文章
译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。
查看英文原文(Underdog shows off a 32MB speech model in new demo)
5:02 AM PST · November 6, 2026
AI companies across the world are racing to make bigger and better models, but the problem is all of them need massive amounts of processing power. On the other end, some researchers and startups are creating smaller models that can run on edge devices, like your phone or computer.
One such startup, Underdog, has released a game built with its new voice model called Woof that’s only 32MB large.
The game features a room with a retro tech theme, a typewriter, and a CRT monitor. There is a desk lamp, which makes a sound when you click its switch. The intro warns users about “intrusive” cloud providers in the town — ostensibly a jab at cloud-based models.
You can interact with the typewriter and music tuner.
We made a game for our new frontier text-to-speech model, Woof. 31.7 MB, so small that it runs in your browser
Play the game on X… all voices are Woof. Defeat Darkcloud for an invite https://t.co/ImUFYDCBQGpic.twitter.com/tthHx7ZGBK
— Underdog AI (@UnderdogAI) November 5, 2026
You can type anything on the typewriter using your keyboard, and the game will speak it back. You can also rotate the tuner to generate different kinds of music.
All of this works within the browser through Underdog’s speech generation model, which, the company says, can run on most devices without any internet connection. The model has 31 million parameters and is open for everyone to download.
While such a small model likely won’t replace bigger models, it may still be enough to drive local assistant experiences or create personalized podcasts.
This demo comes weeks after Underdog launched Bark, a 167 MB speech-to-text model also designed to run on edge devices for transcription and dictation.
Sigil Wen, the founder of Underdog, told TechCrunch that he started building the app because, instead of using notetakers like Granola, he wanted to build his own system with AI models that can run on users’ devices and let them own their data.
“I think we will be able to run most capable models of today on the devices you already own within months. We just want to train the best open models that are American-made, that are open, that are capable and allow people to have private AI,” he said, referring to his manifesto.
In addition to the two voice models mentioned above, Underdog has also released a 27-billion-parameter model that can run on laptops, along with a waitlisted app that the company is positioning as a locally-hosted alternative to Muse and Instinct.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

Ivan covers global consumer tech developments at TechCrunch. He is based out of India and has previously worked at publications including Huffington Post and The Next Web.
You can contact or verify outreach from Ivan by emailing [email protected] or via encrypted message at ivan.42 on Signal.
Loading the next article
Error loading the next article
原文为英文,正文由自动抓取并保留原文。
原文发布于 TechCrunch AI(Underdog shows off a 32MB speech model in new demo),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。