跳到正文
认知
返回认知精选
认知精选科技大约 8 小时前
单一来源当前只有一个可用来源,尚未完成多源交叉核验

Show HN:Subvocal——无声与电脑对话

Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Show HN: Subvocal – Talk to your computer without making noise

434 points by aarushyadav 5 hours ago | hide | past | favorite | 244 comments

Hi HN, we're Aarush and Arnav from The Subvocal Company (YC F26) (https://subvocal.company/). We’re building smart glasses that let you talk to your computer without talking out loud. You can mouth words silently and discreetly and the glasses decode them to text at the speed of speech.

Here's a video of Dave from YC trying it out: https://www.youtube.com/watch?v=9QdX1eObVYI

(Edit: you don't actually have to move your lips or even open your mouth for this to work! Since the RF can see inside your mouth the system can be fine tuned on "internal speech". This isn't the automatic tongue movements that occur when "thinking" words to yourself, but more like mouthing with your lips closed. It's a bit weird to do, but with practice, we have been able to get it down pretty well.)

Arnav and I were studying CS at Georgia Tech at the beginning of the year, with no experience in silent speech or sensing. We were both AI-native builders and power users of Wispr Flow. What bugged us was not being able to use these tools in the library (or an office) or in public. At Stanford’s TreeHacks in February, we decided to build the most technically ridiculous thing we could think of. We’d read about AlterEgo, a silent speech system out of MIT’s Media Lab, and wanted to recreate it in 36 hours. We built a crude prototype, sticking EMG electrodes onto my face:

We got it to kind of work at a letter level in a very janky way. It wasn’t good enough to win the hackathon, but convinced us to go all in. The first time I typed a sentence without anyone noticing, I felt the magic of the technology.

We quickly realized EMG (reading muscle electrical signals) wouldn’t work:

- Signal changes due to sweat

- Electrical interference

- Placement variability

- Most importantly, no one wants to stick electrodes onto their face!!

So we switched to RF and built the glasses you see now. They emit radio waves, similar to Bluetooth, that pass through your face. As they cross the tongue, lips and soft palate, they’re modulated by silent mouthing. We train foundation models on this data to decode these perturbations to text.

Despite prior academic research, getting this to work was a huge engineering challenge. First we built an extremely low noise RF sensing stack. Then we had to place antennas so their fields overlapped maximally with the articulators. Our first attempt on glasses failed completely, so we reverted to old habits and taped antennas on our face:

Since taping antennas consistently in the same spot and angle was hard, we created this weird but effective head rig to hold them in place:

This worked, but obviously wasn’t a product yet. So we worked deep in simulations to design glasses that could hold the antennas.

这演变成了我们现在的设计,天线嵌入在镜框之中。一旦我们确定了存在信号,下一个障碍就是跨用户泛用性。每个人的面部解剖结构,以及由此产生的 RF 信号模式,都存在巨大差异。我们尝试了算法层面的变通方案,但意识到唯一的手段是大幅扩大预训练数据规模。

所以我们就是这么做的。我们在办公室里设立了展位,付费让人戴着我们的眼镜朗读提示词,以此来预训练我们的模型。这产生了一条漂亮的扩展曲线,随着我们扩大数据集规模,错误率迅速下降,跨用户的性能也变得真正引人注目(正如你在上面的视频中所看到的;Dave 只进行了简短的 20 句话校准)。

我们总是听说要从第一性原理(first principles)出发构建事物,但在这次之前从未真正理解。我们曾尝试用自上而下的方法去解决,但彻底失败了。只有当我们把每个组件(天线、电子器件、ML)拆解到其基本原理去理解和优化时,我们才真正让它运行得非常好。

在接下来的几个月中,我们将继续进行研发(R&D),将我们的原型眼镜变成我们网站上那款精美绝伦的产品,并在 2027 年第一季度(Q1 2027)发货。我们还在扩大数据集规模,以改善边缘情况(edge cases)下的性能。

对于任何热爱语音听写却又讨厌在别人面前使用它的人,我们已经开放了首批 500 台设备的预订,最终价格可享受 $200 的折扣。名额所剩不多,请尽快前往 https://subvocal.co/preorder 获取你的设备!

vessenes 3 hours ago | next [–]

aarushyadav 3 hours ago | parent | next [–]

vessenes 3 hours ago | root | parent | next [–]

IHLayman 2 hours ago | root | parent | prev | next [–]

ComputerGuru 26 minutes ago | root | parent | prev | next [–]

d1sxeyes 3 hours ago | prev | next [–]

ciaranmca 3 hours ago | parent | next [–]

joshdavham 3 hours ago | parent | prev | next [–]

sarjann 4 minutes ago | prev | next [–]

genxy 1 hour ago | prev | next [–]

sourpanda 4 hours ago | prev | next [–]

themanmaran 4 hours ago | parent | next [–]

spiderice 2 hours ago | root | parent | next [–]

viraptor 17 minutes ago | root | parent | next [–]

joshgachnang 3 hours ago | root | parent | prev | next [–]

ElijahLynn 3 hours ago | parent | prev | next [–]

_jholland 3 hours ago | parent | prev | next [–]

growt 3 hours ago | prev | next [–]

chfritz 2 hours ago | parent | next [–]

hansonkd 23 minutes ago | root | parent | prev | next [–]

desireco42 3 hours ago | parent | prev | next [–]

SkyPuncher 2 hours ago | prev | next [–]

vakili 10 minutes ago | prev | next [–]

howunfortunate 3 hours ago | prev | next [–]

aarushyadav 3 hours ago | parent | next [–]

genxy 1 hour ago | root | parent | next [–]

skinfaxi 1 hour ago | root | parent | prev | next [–]

atulvi 1 hour ago | prev | next [–]

gh5000 4 hours ago | prev | next [–]

80fj0348i 4 hours ago | parent | next [–]

0x4e 3 hours ago | parent | prev | next [–]

avree 4 hours ago | prev | next [–]

aarushyadav 4 hours ago | parent | next [–]

achandra03 4 hours ago | root | parent | next [–]

gavinray 2 hours ago | root | parent | next [–]

esseph 1 hour ago | root | parent | next [–]

walthamstow 3 hours ago | root | parent | prev | next [–]

achandra03 2 hours ago | root | parent | next [–]

dr_hooo 3 hours ago | root | parent | prev | next [–]

cogythea 4 hours ago | root | parent | prev | next [–]

achandra03 2 hours ago | root | parent | next [–]

kqp 3 hours ago | root | parent | prev | next [–]

1shooner 2 hours ago | root | parent | prev | next [–]

briandw 3 hours ago | root | parent | prev | next [–]

heurist 3 hours ago | root | parent | prev | next [–]

chicagobuss 4 hours ago | root | parent | prev | next [–]

sadilek 2 hours ago | root | parent | next [–]

LorenPechtel 4 hours ago | root | parent | prev | next [–]

sspiff 4 hours ago | parent | prev | next [–]

rickydroll 4 hours ago | root | parent | next [–]

1dom 4 hours ago | root | parent | next [–]

lstodd 4 hours ago | root | parent | prev | next [–]

jbs789 4 hours ago | root | parent | next [–]

rickydroll 2 hours ago | root | parent | next [–]

ronyeh 3 hours ago | parent | prev | next [–]

xbar 4 hours ago | parent | prev | next [–]

mrngld 4 hours ago | root | parent | next [–]

1dom 4 hours ago | root | parent | next [–]

lstodd 4 hours ago | root | parent | prev | next [–]

egeozcan 4 hours ago | parent | prev | next [–]

SkyPuncher 2 hours ago | parent | prev | next [–]

wongarsu 4 hours ago | parent | prev | next [–]

Zambyte 4 hours ago | parent | prev | next [–]

0x4e 3 hours ago | prev | next [–]

sudb 2 hours ago | parent | next [–]

literallyanythi 2 hours ago | parent | prev | next [–]

snitty 1 hour ago | prev | next [–]

vfosnar 56 minutes ago | parent | next [–]

ethanpil 56 minutes ago | parent | prev | next [–]

thisisauserid 4 hours ago | prev | next [–]

40four 4 hours ago | parent | next [–]

frenchie4111 4 hours ago | parent | next [–]

vessenes 3 hours ago | parent | prev | next [–]

Evidence

来源与证据

单一来源 · 1 个信源

当前仅有一个可用来源,尚未完成多源交叉核验。

发布时间
内容处理
AI 翻译 / 摘要,需结合原文核对
内容有误或来源失效?告诉我们

每天早上收到这 5 条

免注册,填邮箱即可;每封邮件都能一键退订。

提交即表示同意按隐私说明保存你的邮箱,只用于发送简报。

或用 RSS 订阅

AI 翻译、摘要和解读仅用于辅助判断;涉及重要决策时,请回到上方来源核实。