Slashdot Mirror
Linux PDA w/Voice Recognition
Hotaine writes "Lernout & Hauspie is working on a Linux PDA with speech regognition built-in. " It can do things like read (and record) emails, find weather, do stock quotes or whatever. Of course, you can't go out and simply buy the vaporware yet, but it has potential.
Why would they patent their version of Linux? The voice recognition is most likely not integrated into the OS...
And I believe they deserve a software patent. People have been working on voice recognition for a long time, and it doesn't work all that great so far... if they did something neat, this is what patents were created to protect...
How much fun would it be if this WAS integrated into the OS? You'd have to give it verbal OS calls instead of common-sense commands... lots of fun.
E.
Does anybody remember a low-budget BBC sci-fi drama from the 1980's called Star Cops? The programme's theme song was by Justin Hayward from the Moody Blues.
Anyway, the main character (a detective called Nathan Spring who was sent out to police a small base on the moon) had this really neat PDA that he called "Box". It was a computer, internet terminal, telephone, diary etc. all rolled into one, and it had a voice interface. And a kind of personality too.
I seem to remember in the end it self-destructed in order to take out some bad guy and save Nathan's life.
Ever since then I've been waiting for these devices to appear. I can't be bothered with handheld computers as they are now, the interface is just too narrowband. I want my own Box.
Consciousness is not what it thinks it is
Thought exists only as an abstraction
Yah! An open-source voice-recogition SDK would be too cool for words. I don't think L&H would go for that, however. Their product line seems focused mainly on shrink-wrap Windows applications. No OSS vibes there :-(
:->
Alternately, there's IBM's ViaVoice SDK, but it's closed-source, and I think it's still in beta. There's also CVoiceControl, but it doesn't look like a very sophisticated implementation. (Then again, I haven't tried either of these). There's still a need for a free-as-speech, smooth-as-glass voice recognition library.
ObOnTopic: Speech control is PERFECT for a PDA-type device. It gives you almost all the flexibility of a CLI, without the overhead of a keyboard. But methinks it'll be tricky to get it running smooth enough so I don't have to repeat commands, or otherwise utter them like two inches from the microphone
iSKUNK!
Erm... I work at L&H, and last tuesday, we bought Dragon... So we're not "generations ahead", Dragon simply doesn't exist anymore ;-)
superblog.org: all your favourite blogs on o
Well, Microsoft owns 7 or 8 % of the stock in L&H. The fact that it runs on linux has to do with the fact that it was attempted on WinCE first, but didn't work, in fact. (I happen to work there, and the guy who led the project told me).
superblog.org: all your favourite blogs on o
I don't know about those products, but for L&H Voice Xpress you need at least 64 Mb of RAM, and preferably even more. So I think those other software packages would need at least as much. Voice recognition isn't easy, ya know ;-)
And it's not running on a Palm, it's running on a Strongarm II.
superblog.org: all your favourite blogs on o
So you pick a few dozen commands that you do often and that can't do major damage without at least some safewords around them.
Voicetyping things into text input boxes isn't as dangerous - so you put some wrong stuff in a paper you're writing, and it's got the usual interjections from editing and real-world interruptions and phone calls. You can edit that later.
Picking context-sensitive sets of things you actually want to talk to your computer about can be tougher...
Bill Stewart
New Fast-Compression-only CPR http://preview.tinyurl.com/dy575ks
Unix-like systems give you a lot of flexibility in letting things talk to each other, including keyboards and their substitutes, processes, etc.,
and they're not picky about what hardware really exists as long as it's not too weird.
The critical issues for porting voice to Unix are things like making sure the scheduler has enough hard real-time support that the voice recognizer doesn't starve, working within a multi-process environment, and dealing with kernel-userspace boundaries. For instance, if your voice recognizer provides context-sensitive vocabularies, how do you keep track of which context(s) belong to which processes, and if they're in the kernel or other protected storage space, how do you pass vocabularies to them efficiently?
But most of that's pretty straightforward, and application level, and can't be worse than adding X Windows or NeWS was, and certainly not as difficult as adding TCP/IP had been.
More importantly, if you want to build voice applications, what kind of services do you think an OS will need to support them? Are they available in Linux? Do you need to add them? Is there anything big you have to get rid of to make them work? Sure, you could junk it all and use a capability-based system like the
Extremely Reliable Operating System
or some microkernel thing. But what do you need?
Bill Stewart
New Fast-Compression-only CPR http://preview.tinyurl.com/dy575ks
A while back, a friend of mine worked at CompUSA (american computer superstore) as a greeter. Every day he would greet people as they came in and help them if they needed it. One day, a woman with sunglasses and a dog came in, and asked if he would help her around the store. John (my friend) was a little confused, as the woman was obviously blind.
They talked for a minute and he asked a fwe questions, trying to figure out who the computer was for. He came to the conclusion that it was for her.
He was helping her around the store when curiosity finally overwhelmed him, and he had to ask her why a blind person was in a computer store buying hardware. As it turned out, everything we do, she did. She was a computer geek. She was even a programmer. I'm not sure whether or not she installed her own hardware:)
She used voice commands for everything. Including writing code.
This point of this diatribe is to point out that while the keyboard and mouse may be fine for you, there are those who need the software to be productive.
As for the exercise, theres always masturbation.
Your own PDA is trained with your 'voice'. Have an external plug on any system, then plug your pda into it. Now you have a voice interface anywhere you go.
The Kruger Dunning explains most post on
I don't like the word 'patent' in the first sentence of the paragraph about this thing. Is it a software patent? I am a bit sceptic about this thing......
Grtz, Jeroen
Secure messaging: http://quickmsg.vreeken.net/
I have used Dragon Naturally Speaking as well as IBM's ViaVoice and neither seem to work too well. I don't think I studder or anything and as far as I can tell I speak pretty clearly. Neither worked well for dictating and they slowed down my computer noticeably. (P200, 48MB) How well can this possibly work on a Palm? Does anyone know what Speech-to-Text engine this uses?
I can see that. When it comes out next year, L&H will change their name to H And L, so that they can call it the 2001 H.A.L. Or, if they chose a more subtle prank, they'll call it the L and H model 1002.
'Holy mirror images, Batman!' -Dick Grayson.
.sig: Now legally binding!
One negative about Voice PDAs, like cellphones, is that you spend time in public talking to yourself while you walk down the street. (Here in my part of San Francisco, half the pedestrians do that because of cellphones, and half because of substance abuse...)
If you add visual displays, e.g. i-Glasses, you're adding extra volume and weight, and you get to walk down the street in Gargoyle mode :-)
Bill Stewart
New Fast-Compression-only CPR http://preview.tinyurl.com/dy575ks
I dont know if this is a true story, a joke, or just a cute little urban legend, but it also regards voice recognition software. This company had just developed a new, really powerful voice recognition program (this was in the old days when DOS when the most commonly used OS), and they were showing off the software at a press conference. The developers encouraged the reporters to yell out commands and watch the software in action. And immedietally, from the back, some wise guy shouted out "FORMAT C:\, RETURN!" The software worked...
Wonder if there could be an analogous action for this new Palm software.
And when they come out with a marketable version, it will be called HAL.
Gonzo
First off, I work for L&H, so if I sound biassed, hey, they pay for my bandwith... But at least I am an engineer and not a salesperson ;-)
;-)
I saw a demo of this thing a couple of weeks ago. Basically what they did was port the Voice Xpress engine to linux, and make it work on a Strongarm II processor. Right now all that was built was a single prototype, and some issues like battery usage are still to be worked out before the thing goes into production. (Estimate : end of this year)
What was also done, was building in the L&H Realspeak synthesized voice. This allows real dialogues to be held with this device. For instance :
-Check mail !
-You have five new messages
-Summarize !
-Message one, sender is J. Random Hacker, Subject : kernel hack. Message two, sender is Bill Gates, Subject is Make money Fast...
-Read the first message !
-Hi there... I just finished a patch for the latest kernel update... blablabla...
-Reply !
You could then use this to dictate an e-mail (because it uses a full Voice Xpress engine), or order books on Amazon (If you wanted to shop there) using WAP. All in all, a pretty nifty device. I want one myself
And no, it was not vapourware, it really works : with the VX engine you can already get 95% accuracy if you want to, with a large vocabulary. The smaller the vocabulary, the better the recognition. How many commands do you need to read your e-mail ? You do the math.
Oh, yeah, it also has a stylus and screen that fold away into it, just in case you really wanted to read or write stuff.
superblog.org: all your favourite blogs on o
In general, taking an operating system and strapping a new paradigm to it is a Bad Idea.
When Microsoft took DOS, a single-user operating system, and added Windows to it, they ended up with a really terrible hack that didn't run either, up until about version 3.1.
When Microsoft later took Windows and added pen support (not CE, but Pen Windows--read your history!), it flopped. Windows wasn't suitable.
When they attempted to port a subset of NT to pen-based devices, they got CE, which has underperformed.
On the other hand...
When Apple abandoned the Apple II and designed the Mac as a graphically-operated machine from the ground up, they got a great platform.
When Palm threw out common notions of a shrunken PC and designed their device from scratch, they won.
My point is that a good voice-controlled system needs to be designed as a good voice-controlled system. Strapping it on after the fact won't cut it.
Me: "p-w-d"
Computer: "slash"
Me: "l-s"
Computer: "slash u-s-r, slash h-o-m-e, slash b-i-n, slash b-o-o-t, slash e-t-c..."
Me: "cd slash e-t-c"
Computer: *silent*
Me: "vi profile"
computer: (reads file contents)
Me: "quit"
computer: "error"
Me: "exit"
computer: "error"
Me: "wq"
computer: "file is read only"
Me: "q!"
That's my kinda computing
-josh