I've started a Blog!!!

Toggle sidebar Toggle sidebar
Schadenfroh

Schadenfroh

Elite Member
Mar 8, 2003
38,416
4
0
Herbivore.. Insanely cool P2P project
Looks like some friends of mine at Cornell are working on something really cool -- a project called Herbivore.
In a nutshell, it's a Peer to Peer system that ensures privacy. Nifty.
Read more about it at www.herbivore.info
And look at their whitepaper here.
posted @ 12:10 PM | Feedback (0)

GM-retreat -- thoughts and concerns
So yesterday was the GM retreat, and there was some cool research being presented -- I especially like some of the new drive-by-wire stuff coming out, as well as the redundant "Flex-Ray" bus. I also found out that GM is opening up an embedded software research lab in Bangalore, India -- somehow I get the impression that this will be a software sweat shop of sorts, and they'll get 1/5th the salary of their US counterparts. I hope for the sake of the people that end up working there that that doesn't happen, but I just didn't get good vibes about it. People over here in the US (Michigan) research labs seem somewhat concerned about losing their jobs within two years to people in India, and Bangalore is having it's own dot-com boom of sorts now -- I heard some figure that if you lose a job in Bangalore, you will get rehired in two-and-a-half weeks.

As for the dialogue systems bit -- GM doesn't really seem to know what they're doing in the realm of them -- One person, who shall remain nameless in this blog, gave a rather pessimistic outlook on them, and only showed an example of a very very limited directed dialog scenario. I politely proved her wrong after showing her Ariadne, and explained that having a fast CPU and a hard drive for a dialogue manager is a necessity, that these "embedded directed dialogs" simply do not make the users happy. I took this as an insult, as GM has supposedly had access to work on Ariadne for well over 6 months now, and it seems like not one single person there invested the time to even write a dialog scenario for it! Pathetic. I was also very annoyed by the amount of resistance GM put to the idea of having roughly the equivilant of an XBox's hardware with more RAM as a platform within the car. They went and whined about how any CPU that costs more than $10 is too expensive for them), that only one manufacturer (Honda) has thought about putting a hard drive in the car, and that it's too hard to deal with thermal requirements to put a hard drive in a car that will last the temperatures and vibrations. Okay, enough flaming of them but if you are going to keep whining about the engineering challenges of sticking essentially a $200-$250 PC in the car is, then I really have to question how much I want to work with them in the future.

Bosch's researchers, were totally receptive to me when I presented Ariadne to their Palo Alto research office (perhaps it helps that 6 of them have PhD's in Natural Language Understanding), asked me very very relevant questions than were *dialogue system* related, not "Can we engineer this to really fit into a car" related, and they really knew their stuff. I'm getting out this Ariadne beta by the 17th of October... is it any wonder that I'll be more receptive to Bosch than GM? As far as Bosch is concerned, their Navigation system needs 2- 5 years more to mature, and the computing requirements to them are no big deal. And I have to say I agree with them. If the hardware costs are even $500, then it would be a $2000 option in cars, if the hardware costs are $250, then it would be a $1000 option. Current Navigation systems sell for a $2000-$4000 premium, so I don't see this as an unreasonable extra expense. Bosch doesn't either... but GM can't even convince their purchasing department to let their headunits do both MP3/WMA decoding (they only will do MP3 decoding) because an extra $0.12 per headunit in licensing fees to Microsoft. Again, pathetic.

Supposedly Bill Gates has been over to GM about 20 times regarding in-car computing, and GM's managers never listen to him, either.

Perhaps this should be a wake up call for me to go to where I'm received well, and the bean counters don't rule over everything. I thought GM would improve a lot because of Bob Lutz, but it seems like they still have a long way to go. If anyone here knows Bill, feel free to direct him here and let him know he isn't alone at being frustrated with GM. The Germans may once again lead the way, while internal fighting keeps happening at GM which serves to do nothing but stifle innovation. Rather than arguing with me, they could have said "this is a good idea, why don't we prototype a good hardware platform to support this that could exist in the car environment?"

Here's a hint GM -- everything is expensive at first in the technology world, and gets cheaper over time. Two years ago, Ariadne would only run on machines that cost $3000+. Now I'm saying it could run on a $250 computer. Wake up General Motors.

posted @ 11:44 AM | Feedback (0)

Wednesday, October 01, 2003 #

Microsoft Speech SDK 6.0 Beta 3 problems
Ok, I've confirmed this problem -- the SAPI Server (SAPISVR.exe) loses responsiveness after running for a while (30 min+) -- It still will recognize, but not with the responsiveness when just launched. Time to see where I can go file a bug report...
posted @ 9:12 PM | Feedback (0)

Wow -- Go Optimizations, and Thank You, Visual Studio 2003
Wow, so I spent the last few days doing lots of optimizations for Ariadne, and got the RAM usage down to 17-19mb!! This is down from 32 Mb. A substantial difference, if I do say so myself, and I finally got it to compile with lots of aggressive optimizations turned on (previously, Ariadne would compile OK, but crash when running if I turned too many optimizations on). People with Pentium 4's will love this program :)

Also kudos to Microsoft's new 6.1 recognizer -- it is still a bit buggy (sometimes seems to stall, occasionally crashes, but runs phenemonally well, especially when trained. Ok, I cheated. I trained the recognizer for Ariadne on my own system. Ariadne recognizes pretty well without the training, but training helps! I think i have the build that I want to be the beta build finished now. Just have to write the documentation now...

Visual Studio 2003 totally rocks -- I swear this new compiler for VC++ is much smarter, and everything just runs faster than VS 2002. FYI, Ariadne is a totally C++/Win32 app -- NOT using any managed code bits. I prefer the extra speed for an app like this. As for the Microsoft Speech (SAPI Server) SDK 6 beta 3 -- I am seeing the following bugs/issues:

1. The recognizer almost seems like it falls asleep when I haven't said anything for a while and takes a few phrases to wake back up. Anyone else experience this?
2. Sometimes, it seems like the recognizer launches and doesn't see the microphone... really frustrating, didn't happen in beta 2.
3. The recognizer appears to get less responsive after a lot of use.
4. It crashed for me during training once. I had none of these issues with Beta 2.

My system is a P4-2Ghz running Win2k Server SP4, and my other computer is an Athlon-850 running WinXP SP1 and Win2003 Server. Let me know if any of you see any of the same problems with your speech SDK applications.

posted @ 8:50 PM | Feedback (0)

Use of Ariadne in Home Entertainment and Cars
I am posting this rather than replying so more people see this --

re: First Post! 10/1/2003 3:35 PM Kartal Guner Yes, I saw the video. It would be really cool if this could be a plug in for something like http://myhtpc.net . Very useful if used in car environment, to keep your eyes on the road.

-------------------

Actually, I already have my eyes set on both the Windows XP-based media centers and car companies as well. Bosch's HMI (Human Machine Interface) group in Palo Alto loved my presentation of it this summer, but their navigation system isn't ready yet -- it'll be about 2-5 years away. If, however, someone on Microsoft's AutoPC/WinCE Automotive group sees this -- please please give your navigation system good external interfaces -- and give me a call! Same with the MapPoint group -- MapPoint is in fact what I am trying to interface Ariadne with now, we'll see how well it works.

posted @ 4:09 PM | Feedback (0)

Ariadne Beta coming soon (within 2 Weeks)
I'm preparing Ariadne for a beta trial now -- I have to write some documentation and an installer for it! I think I'll have something posted within two weeks. Will post it here when it's ready! Let me know if you liked the video!
posted @ 3:10 PM | Feedback (2)

SpeechTek 2003
A whole bunch of friends of mine went off to this -- be sure to check out the OpenSALT specification being developed at CMU!
posted @ 2:55 PM | Feedback (0)

Notes on the Ariadne Video
That video link below was from 2/2002, since then, we've made lots of changes. Ariadne works with the Microsoft Speech SDK 6.0 beta 3 quite nicely now, and we are utilizing Phoenix rather than Soup Grammars, and utilizing Microsoft's Speech Recognition engine rather than JANUS, which is a CMU/University of Karlsruhe research project. Here's a screenshot of the newer builds of Ariadne
posted @ 2:01 PM | Feedback (0)

GM Collaborative Research Lab Retreat
For all of us at CMU who are funded by General Motors, we have a retreat this Friday!! This is going to be a great opportunity to get to know some of GM's top researchers, and learn about all sorts of cool research going on over here! Free Food and Free Knowledge! Can't beat that!
posted @ 12:05 PM | Feedback (0)

First Post!
Alright, so today I'm starting a new blog relating to Speech Technology and Dialogue Systems. I'm a senior at Carnegie Mellon University in Computer Science, and I've spent about two years working on dialogue systems here.

This blog will contain news on current speech technology and other technologies of interest to me (mobile devices, web services, automotive technologies, telematics, and more), my take on them, as well as my own trials and tribulations with development of a dialogue manager we call Ariadne.

Ariadne is a dialogue manager platform of sorts which I've worked on with Matthias Denecke, who did his PhD in Computational Linguistics, and Ariadne was part of his PhD thesis. You can see a video demonstration of Ariadne here:

(right-click the link, save as -- it doesn't stream. You need the DivX codec from DivX.com

I gave a talk on Ariadne at the Software Development Forum Emerging Technologies SIG in August -- you can see information on that here.

posted @ 10:59 AM | Feedback (6)
Click to expand...



FOR THE LOVE OF GOD PLEASE SOMEONE READ THIS AND MAKE CLIFF NOTES
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom