Registering and logging in removes this ad.
Registering and logging in removes this ad.
laptop reccomendation
Submitted by voicedictation on Mon, 09/19/2005 - 00:20.
I am in the market to buy a laptop computer for use with DNS, but I haven't begun my research/search for it yet.
I'd like a high performance laptop that:
- is relatively lightweight to carry
- has a fast(ish) CPU
- can run Windows XP w/SP2
- runs NatSpeak Professional edition simultaneously with several other applications, such as MS IE, MS Outlook, MS Word
- comes with a great sound card or integrated chipset that is compatible with NatSpeak
- has about 1 GB of RAM (?) for good performance
- doesn't cost an arm and a leg
If you also can recommend a headset/microphone that is known to work with the laptop you recommend, I'd love to hear about that as well.
Thank you in advance. John
voicedictation, I personally
voicedictation,
I personally use an IBMT40 with 1GB array of a 1.6GHz CPU. Dragon is a tad slower than on my high powered desktop workstation. For me it is not annoying. There are more tea series IBM ThinkPads that may have a slight speed boost. I personally use the built in soundcard (a well shielded SoundMax) with my Sennheiser Headset.
--
Martin Markoe, eMicrophones, Inc.
The best microphones for Speech Recognition
See us at: http://www.eMicrophones.com/index.asp
Read, "Key Steps to High Speech Recognition Accuracy" at:
http://www.emicrophones.com/docDetails.asp?DocumentID=38
Martin: Thank you for your
Martin:
Thank you for your information about the IBM Thinkpad T series. That looks like a good system for NaturallySpeaking, and it is lightweight, too. It is a little more expensive than the HP Pavillion dv4000 notebooks.
Everyone:
Does anyone have any first-hand experience using NaturallySpeaking on an HP Pavillion dv4000 notebook?
If anyone has any other recommendations for specific notebooks for use with NaturallySpeaking Pro, I'd love to hear.
My next question is about maximum amount of memory. Right now, with NatSpeak 7.3 Pro, I get along fine with 1GB RAM on my desktop computer, running mutliple apps (MS IE, MS Word, Outlook, etc.). If I buy a laptop that will NOT hold more than 1 GB RAM, do you think future upgrades of Naturally Speaking Pro in the next few years will require more memory? (If I am limited to only 1 GB RAM on the new laptop, if future releases of NaturallySpeaking require more memory than the current release, then I'd get sluggish performance. (I don't want to upgrade the laptop for a long time.)
Thanks, John
voicedictation wrote: My
My next question is about maximum amount of memory. Right now, with NatSpeak 7.3 Pro, I get along fine with 1GB RAM on my desktop computer, running mutliple apps (MS IE, MS Word, Outlook, etc.). If I buy a laptop that will NOT hold more than 1 GB RAM, do you think future upgrades of Naturally Speaking Pro in the next few years will require more memory? (If I am limited to only 1 GB RAM on the new laptop, if future releases of NaturallySpeaking require more memory than the current release, then I'd get sluggish performance. (I don't want to upgrade the laptop for a long time.)
Thanks, John
John,
Regardless of what anyone may say, DNS can use up to a maximum of 256 MB of RAM. It cannot use any more RAM than that, and it does not need any more RAM than that. In fact, DNS seldom uses more than 100 to 140 MB of RAM regardless of the user's configuration or the language/vocabulary selected. It is highly unlikely, primarily because of performance issues and speed, that DNS will access more than 256 MB of RAM even in the new 64-bit operating systems.
The real function of RAM relative to DNS is in terms of avoiding swapping out to the pagefile/swapfile. Having more than 512 MB of RAM increases the Windows system performance, but has no actual impact directly on DNS. Because Windows system performance is improved and the ability to run more applications simultaneously without swapping out to the pagefile, this has a natural effect on improving DNS performance. However, this effect is indirect. In other words, if all you ever ran on your system was DNS and the only application that you ever used was DragonPad, there would be no significant difference in the performance or speed of DNS whether you had 512 MB RAM, 1 GB of RAM or 2 GB of RAM. Having more RAM only improves system performance. Since all of DNS is loaded into RAM at runtime, adding more RAM has absolutely no impact directly on DNS speed or accuracy.
The bottom line is, relative to your question, if you get 1 GB of RAM on your laptop, you will have sufficient RAM to run DNS for likely several versions in the future, at least. More than 1 GB of RAM is a matter of personal choice and how hard you want to run your system (i.e., how many applications you want to run simultaneously).
Chuck Runquist
former DNS/SDK product manager for technical solutions at L&H
Here's some information on
Here's some information on how to determine if the amount of memory you have is suitable for the applications you're running http://speechwiki.org/SR/HowToDetermineAmountOfMemoryNeeded.html
-- Scott W
Speech resources at SpeechWiki
Products at Applied Recognition
http://tinyurl.com/9h4ss (long url replaced by admin)
I have the notebook described on the Web site referenced above. I have Dragon NaturallySpeaking version 5 and ViaVoice version 10 installed and use them both interchangeably. I use a moderately priced headset (I work in a quiet environment) connected directly to the notebook sound card. I am well satisfied with the results.
Charles
Have a look at
Have a look at http://www.speechcomputing.com/node/220. There is a full discussion on computers and their the specifications which are satisfactory for SR.
Quentin
Hello again, Just curious.
Hello again,
Just curious. How does a larger L2 cache affect NatSpeak's performance? For example, assuming 1 GB of RAM is adequate, if the only variable that changed is the size of the L2cache, would more L2 cache speed up NatSpeak?
Thanks, John.
voicedictation wrote:Hello
Hello again,
Just curious. How does a larger L2 cache affect NatSpeak's performance? For example, assuming 1 GB of RAM is adequate, if the only variable that changed is the size of the L2cache, would more L2 cache speed up NatSpeak?
Thanks, John.
John,
Your question is a little bit like asking whether a thoroughbred is faster than a quarter horse. The answer is, it depends on conditions.
First, L2 cache is only used by the CPU. It is not accessible by applications.
Second, under most conditions increasing the size of the L2 cache will increase the speed and performance of CPU processing where multiple applications are requesting CPU time simultaneously. However, its overall effect on the speed of DNS is largely relative. For example, I have 2 identical systems. The only difference is that one is a P4 3.2 GHz with 1 MB of L2 cache. The other is a P4 3.2 GHz Extreme Edition with 1 MB of L2 cache and 2 MB of L3 cache. The first system cost me approximately $980 to build. The second system cost me approximately $1800 to build because the CPU came in at approximately $989. Under normal conditions, the first system running DNS and Microsoft Word only is not significantly slower than DNS and Microsoft Word running under the same conditions on the second system. However, if I am running several CPU intensive applications on both systems along with DNS, then there is a slight, but still noticeable reduction in the speed of DNS on the first system vs. the second system. Nevertheless, in the long run the difference is negligible. Therefore, L2 cache may or may not improve the speed of DNS.
The point is that the current standard is 1 MB of L2 cache and that is probably desirable as a minimum for optimal performance with DNS. However, don't bank on L2 cache size because that is not the only performance factor, and increasing the L2 cache size takes more away from your pocketbook than it necessarily adds to the speed and performance of DNS. There was a significant difference in performance when the standard for L2 cache was 512 MB. However, if I am correct, and anyone who knows can correct me if I'm wrong, I believe the current AMD processors still run a 512 MB L2 cache. Yet their performance with regard to DNS is as good as or better than P4's running 1 MB of L2 cache.
In the short run, a quarter horse is faster than a thoroughbred. In the long run a thoroughbred can maintain that speed and performance longer than a quarter horse. The question is how much are you willing to pay for either. Or, how much are you willing to pay for a CPU with a larger L2 cache size. The cost-benefit ratio doesn't necessarily benefit you or DNS, even though, under some conditions, it can, but only indirectly. In the long run, you would be better off from a cost-benefit standpoint buying a dual core CPU, which is significantly cheaper and faster even without hyper threading, than my P4 Extreme Edition when using DNS under most all conditions. Remember that L2 cache is 10 to 20 times faster than RAM with an equivalent cost differential. To my way of thinking, my new dual core P4 3.0 GHz was a better investment (i.e., considerably cheaper), and DNS runs faster on it.
Chuck
Good, balanced answer,
Good, balanced answer, Chuck. I would emphasize that its impossible to ascertain the effect of a particular cache unless you're a CPU developer who can actually generate experimental data varying the cache size.
You have to assume that the CPU makers have good reasons for the particular cache sizes and allocations they offer commercially. However, the best way to assess a particular CPU, with all its features, for running a particular application is to run that application with a variety of CPU architectures.
You virtually can't do that anymore -- first of all, CPU development has stagnated in the physics of decreasing die size and its comcomitant nonlinear increases in electrical leakage/heat wastage. Secondly, CPU enthusiasts are mostly either gamers or scientists/engineers/financial analysts. Both groups have specific operational interests that don't include SR -- nowadays no one includes SR apps in benchmarks.
So strictly speaking, its hard to say which CPUs run SR fastest -- except that its not really an issue any longer. Both Intel and AMD CPUs have more than enough power to run SR faster than ever imagined before.
The main reason CPU makers are increasing cache size is that its something they can do that has a slight impact on performance, especially on multitasking. They simply can't get their CPUs to run any faster except at a rate roughly an order of magnitude slower than previously. What can they do? They can put more cores on the die, they can increase cache size, etc. But they can't whip them puppies to make them run any faster.
Frankly, all the high end CPUs run SR pretty fast these days. What we really need is a fundamental way to re-engineer how SR works, and that technology is no where in sight.
Whats to be done? Learn how to use the available software more effectively, because for the next decade you are likely to see decreasing marginal benefits from each CPU and software change.
Bruce
PS: Does anyone know why the messages in this thread extend beyond the normal screen size so that I have to use the horizontal scroll bar to read them? This is an EXTRAORDINARILY AGGRAVATING aberration that makes it harder for some folks to follow the discussion.
If Admin is reading this, could he please fix the site so that this doesn't happen again? Maybe charge whoever's post causes this to happen a penny for everytime a reader has to scroll in more than one dimension
BruceCyr wrote:You virtually
You virtually can't do that anymore -- first of all, CPU development has stagnated in the physics of decreasing die size and its comcomitant nonlinear increases in electrical leakage/heat wastage. Secondly, CPU enthusiasts are mostly either gamers or scientists/engineers/financial analysts. Both groups have specific operational interests that don't include SR -- nowadays no one includes SR apps in benchmarks.
So strictly speaking, its hard to say which CPUs run SR fastest -- except that its not really an issue any longer. Both Intel and AMD CPUs have more than enough power to run SR faster than ever imagined before.
Bruce
Bruce,
Generally, this part of your post is correct. However, having been a part of these discussions at L&H, I know that Intel has always tested every version of DNS, and before that L&H VoiceXpress, with their CPUs and motherboards. They even have a specially designed C++ software compiler for target recompiling applications on Intel systems. I broke 2 legs and an alarm
trying to get L&H to do a target recompile of both VoiceXpress and DNS. Never came to pass. However, I'm still trying to get ScanSoft to do it. Maybe one day 
they will.
Interestingly enough, and many people probably don't remember, but the introduction of MMX technology was something that speech recognition could really get its teeth into. Intel added an 80 function instruction set, built primarily into the FPU, that was specifically designed to enhance speech recognition. The primary reason for this was that speech recognition makes heavy use (i.e., Markov modeling) of FPU cycles. Therefore the introduction of these additional 80 function instruction sets and the advent of MMX technology significantly benefited speech recognition, even though they were initially and primarily designed to benefit gamers. That was an interesting time and allowed SR can make some big inroads in speed and accuracy.
So, even though Intel and AMD are not benchmarking SR products publicly, at least Intel is still interested how well speech recognition performs on their platforms. A core engineer for Intel, who happens to live down the road from me, uses DNS all the time and is always testing and reporting on how well it works or doesn't work with every new change in the technology. He has told me some interesting stories about how he thinks SR, and particularly DNS, can take advantage of the combined Extreme Edition P4's dual core and hyper threading technology. Of course, I'm begging him to get me a dual core/hyper threading Extreme Edition P4 without having to pay the $1400 that it currently costs (i.e. of course give or take a couple hundred in change). So far it's no joy
Chuck
Oh, yeah, that's good
Oh, yeah, that's good background for sure! The reason DNS got included in the SYSMark benchmark was because of its use of those MMX instructions -- its performance showed off the comparative power of Intel CPUs.
But we're talking from different perspectives. I should have said that from the user's point of view there is no longer any SR benchmark data available because SYSMark became so biased towards Intel that it fell out of favor with the enthusiast crowd, most of whom were also AMD fans. So basically benchmarking sites have either dried up or else focused exclusively on game-based benchmarks, which are their primary interest anyway.
Nowadays the only public benchmarks that I know of which include SR performance are those that tech-reports.com uses, i.e., Sphinx 4 SR performance. Formerly, every performance site, including those in Korea, Japan, Finland, Poland, Russia, etc. as well as the USA used SYSMark, and generally broke out the individual app results, which included DNS 4(?). FWIW, the Korean, Finnish and Polish sites I monitored had the most professional and exacting engineeering approaches to benchmarking -- and I can't even read those languages!
It would be nice if we could persuade Intel and AMD to release more data -- and have them do systematic experimental trials including controls on memory, hard drive speed, cache, etc., so that we could just look up which CPU runs DNS or VV fastest
Bruce
PS: Maybe a bunch of us could do an on-the-fly-group buy from your neighbor
Quoting Bruce: "PS: Does
Quoting Bruce:
"PS: Does anyone know why the messages in this thread extend beyond the normal screen size so that I have to use the horizontal scroll bar to read them? This is an EXTRAORDINARILY AGGRAVATING aberration that makes it harder for some folks to follow the discussion."
I do not experience this phenomenon. The messages extend a little more than three quarters of the way across the
screen. I'm using a 17 inch flat-panel monitor at a resolution of 1280 x 1020 pixels.
Charles
It depends on the size of
It depends on the size of your resolution.
If, like me, you are using Mozzilla, and increasing the size of the resolution, there is no doubt that some of the messages disappear to the side of the screen. It is just a matter of reducing your resolution to fit the screen.
Quentin
P.S. hoHow did we get off the subject again!
Bit of a mystery. I'm
Bit of a mystery. I'm running same resolution at 19 inch. The phenomenon has nothing to do with my screen resolution -- I get the same thing at 640x480 and every stop in between.
This page is fine, but page 1 is messed up as described. It begins with the initial post by voicedictation -- wonder if he used any html coding?
I wonder how many other people do/don't see this effect. BTW, do you also not have the off-screen extension on the previous page?
Bruce
PS: I'm running Mozilla 1.07 (hope everyone has upgraded!), but it was the same yesterday in 1.06. What browser are you running?
BruceCyr wrote:PS: Does
PS: Does anyone know why the messages in this thread extend beyond the normal screen size so that I have to use the horizontal scroll bar to read them? This is an EXTRAORDINARILY AGGRAVATING aberration that makes it harder for some folks to follow the discussion.
If Admin is reading this, could he please fix the site so that this doesn't happen again? Maybe charge whoever's post causes this to happen a penny for everytime a reader has to scroll in more than one dimension :-)
It's not the site really. It's the way Firefox is interpreting the screen. I didn't have this happen until I upgraded to Firefox 1.0.7 today then it went all over the place. Internet Explorer didn't seem to have this problem. Subsequent attempts with Firefox seem to be ok now.
I can read the site and this thread on my Pocket PC (tiny screen) and it's formatted correctly even there.
If I'm offbase and you notice a pattern, please send feedback using the Contact Us form and I'll see what I can do.
I'm getting the same
I'm getting the same off-screen pattern with both your post and that of Bruce immediately before it.
Using Firefox, use the Ctrl+ scroller to adjust the screen to fit the entire message.
Quentin
Looks like the very long URL
Looks like the very long URL in Charles' post is why Bruce had that problem. I had the same, and just for pg. 1 also. I typically use the HTML tags provided to embed a URL when they are that big.
My original curiosity about L2 was in that Cyrix, now Via, CPUs run very cool I think because they have sometimes as little as 64k of L2 cache. I think that would be too weak. I am mentioning that partly because there was a post a few weeks ago asking if a mini-itx computer would be suitable for SR, and although I haven't tried SR with one, I suspect it wouldn't be so great if you used a Via mini-ITX with their embedded CPU. (Apparently it's time for embedding myself too.) I have a mini-ITX 800 MHZ board here, and it performs about like a 400 mhz P2, once you start doing anything sustained with it.
Good night, Tad
voicedictation wrote:Hello
Hello again,
Just curious. How does a larger L2 cache affect NatSpeak's performance? For example, assuming 1 GB of RAM is adequate, if the only variable that changed is the size of the L2cache, would more L2 cache speed up NatSpeak?
Thanks, John.
John,
After reading all the replies to your question, it seems to me that the whole matter resolves itself down to cache as cache can.
Charles