Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
How the NSA converts spoken words into searchable text (firstlook.org)
189 points by jbegley on May 5, 2015 | hide | past | favorite | 22 comments


I've worked on similar problemsss starting with Microsoft's speech SDK back in the 90's. This is more of a pain the the ass than it first appears to be. I was secretly excited that maybe NSA has cracked a difficult nut here. (NDA be damned, I'd like to see this in the commercial market)

While the whole spying thing is disconcerting, I wonder how much cool tech they have that could be put to commercial use. wondert if theres a path for a government agency such as NSA to share tech breakthroughs with the public sector. Nasa does this to some degree...


https://www.nsa.gov/research/tech_transfer/

Most government agencies that do research have a technology transfer program. Searching for "<agency> technology transfer" gets you to the correct place.


This is cool: http://m.nsa.gov/research/_files/tech_transfers/nsa_technolo...

Btw. the mobile and desktop version of the NSA's webdesign could take a facelift.


Rationally, I know that any PDF could infect my host, and that the NSA doesn't need me to download it from them to get an intrusive payload. Still, I am not opening that PDF.


This is likely one of the the best responses Ive gotten on a HN comment! They really do have some sweet tech you can potentially license from them! I wish I could upvote more than one. Thanks for the info adestefan


I don't understand how some of these merit patenting. Page 11, for instance, describes a patent on Comparing Voice Signals which seems to be an extremely simple form of data augmentation - playing an audio file backwards and rearranging chunks of it. That's hardly groundbreaking.


Can you drop me an email? I have a project to show you. :)


I thought that all NASA technology was open to the public... Isn't that how SpaceX started?


One of the quirks of tech transfer offices is that they have emerged from government agencies and universities because it turns out that making technology and patents freely available doesn't do much for commercial adoption.


It's not... you're probably thinking of the fact that anything NASA releases is not copyrightable. This doesn't imply, however, that they must release everything.


Anything sufficiently sensitive or representing a US advantage over other nations is ITAR controlled.


NSA, not NASA.


FYI: I was replying to this part of the comment: "Nasa does this to some degree..."


I rather assumed the tech transfer for this went in the opposite direction: that a reason Google, and Microsoft, and Apple were developing, and providing for free, voice recognition on consumer devices was that they wanted to get the technology to a point where they could sell, or license, it to the NSA.


That's a very strange way of looking at it. Those companies would want to sell such a service or product to anybody and everybody, including the NSA. You might as well say that the reason Dole grows pineapples is that they want to sell those pineapples to the NSA.


You could use that analogy, but only if the NSA really really liked pineapples. They really really like spying on people and therefore like the technologies that allow them to do so. And they have a lot of money and power that would appeal to a company that might want to dabble in the gray areas of legality with regards to privacy.

You might argue that Facebook works to perfect facial recognition and identification technologies to sell them to the NSA, along with that nice full database of people and their pictures. That would make more sense than pineapples.


I think the pineapple thing could be a major insight. After Microsoft, the NSA's famous mission statement was to put a pineapple on every desk.

The DoD explicitly want to "deter war and provide security for our pineapples" and NASA infamously put a test pineapple into space in 1963, long before a test pilot.

Of course we must not forget the clearest endorsement of pineapple in American Life, in the most famous text of all, where Jefferson (himself secretly addicted to its juices) said "Life, Liberty and the pursuit of Pineapples"


A reason, not the reason. I don't imagine these companies discount the sort of money that the NSA seems willing to throw at these sorts of problems. A sum that will not be gained from what is, given the quality of the microphones and the maturity of the algorithms, a novelty item in other respects at the moment.

If the technology matures, yeah, you'd want to sell it to everyone. (Actually you'd want to sell different forms of the technology to everyone, or at least different licenses, so that you could bleed them of as much money as possible.) But in its current form, and perhaps for some years after depending on how much of an edge you'd think it gave you in the market, you'd want to sell to rich customers who have a strategic interest in the technology and can give you some sort of decent return on your development costs.



There is an easy API for this now: http://Clarify.io

We use it to allow our customers to search their calls and voicemail messages.


I just assumed they had dragon naturally speaking running on a pentium and windows 95


The floating point bug prevented that project from succeeding.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: