Sphinx by itself is a terrible answer to this problem, unfortunately. The code is free, but good luck finding an appropriate model. Worse, you'll need to train a speaker-dependent model to get any usable results, and this is a VERY non-trivial task with Sphinx tools in the state that they are. I spent several years getting paid to adapt Sphinx for commercial purposes and while it's great for some things, I can say with confidence that it is not the tool you're looking for.
You know what works? Dragon. Hate to say it, but the commercial products here have a gigantic edge on the competition.
That said, I'd love to see someone come up with an open source speaker-dependent model training system that's friendly enough for app developers (not speech researchers) to roll into projects. I think this is a big open door for contribution to the community. Sphinx isn't the best thing going, but it's certainly usable, and if a real product came into being I'm sure all the speech wonks would start coming out of the woodwork to improve the algorithms.
Algorithm revised in light of real-world performance constraints! Read all about it!
Seriously, we just rewrote a tree (that runs in a high traffic serving environment) this month at work because it wasn't streamlined just right to take full advantage of the underlying architecture. No one will write a paper about it.
Even if generation 1 electric cars are a mess, SOMEONE has to make them viable enough and sexy enough to get the market moving in that direction. As soon as electric cars start becoming a significant chunk of the automobile market we'll see battery, motor, and material research go through the roof. This has already started, and I think it would be hard to argue that Tesla hasn't played a big role in swaying the general perception of electric cars from "slow ugly thing that hippies drive" to Serious Business.
Aptera is awesome in their own right, and you're right that their design pushes the envelope a lot further. Hopefully gen-2 mass market EVs will go more in that direction. I just don't think anyone should downplay the importance of Tesla (etc.) making the generation of cars that bridges the gap between what we drive now and what we'll drive once electric cars are pervasive.
Part of my password stays the same; part changes. The part that changes, I write down on a post-it (literally). The part that stays the same is memorized.
In practice, I have a sandwich, xxxxxYYYYzzzzz, where x and z are constant and Y changes to meet the needs. This is also how I customize for different applications, e.g. my slashdot password might be xxxx/.zzzzz, my bank password xxxx$$zzzz, etc.
It works. It's pretty safe and easy compared to the alternatives.
"FB can't possibly keep backups of every state of your profile"
Wrong, for two reasons: (1) as sites and traffic grow, old data becomes a smaller and smaller slice of the total storage pie, and (2) storage just keeps getting cheaper. Also... just how much space do you thing profile information takes up, anyway? I'm guessing photos and activity logs are the bulk of their storage costs.
Sure,.net would have been way better if it embraced Linux and welcomed re-implementation from the beginning. It would have been more widely adopted, there would be more open source VMs and IDEs, etc.
Why? Because people love free stuff. Good for the ecosystem doesn't necessarily mean good for Microsoft. No one wants to pay for a Windows license for each machine in their datacenter. How many Solaris licenses did Java sell for Sun?
The way I see it,.net was yet another decent attempt at forcing Windows lock-in (and keeping already-locked-in partners happy and productive).
Meanwhile, somewhere off the coast of England, a colossal killbot picks up the news headline from a radio transmission and pauses to reconsider its mission of vengeance.
oil futures aren't exactly a junk stock. good luck making an impact on those prices with a spam email campaign. there's a reason they usually target little stuff you've never heard of.
i'd never heard the term "model selection," so thanks for pointing that out. it looks like there really is some good literature to read on the subject.
the process described by the model selection sites i skimmed still doesn't adress what i was getting at, though. "choosing a model from a set of potential models" is only conceivable when your set of potential models (and set of variables to potentially be modeled) is well bounded.
to put it another way, take the smartest model choosing algorithm you can find, hand it a pile of data, and say "what do you make of that, smart guy?" i'm willing to bet that the answer is going to be along the lines of "wtf?" unless there is some sort of context or metadata provided along with the data to give the algorithm a hint of what it's looking for. am i looking for covariance between scalar values among regularly organized groups? am i looking for white rabbits in the image data from a camera? is this ascii or ebcdic or 8-bit PCM data? you can argue that these questions are trivial, that no algorithm can be *that* general, but that is precisely my point: all known algorithms require significant narrowing down of the problem space by human hands before they can begin to produce useful output.
if you had an algorithm that took *truly* semantics-free data in one end and spit models of regularly occuring features out the other end, you'd be halfway to general AI.
the difference is that brains create new theories and models to describe data, whereas this article specifically talks about avoiding the need to make new theories to describe data. we still have no AI that can create theories and models and semantics on its own. i agree that when that happens, we'll have something exciting and new, but it hasn't happened yet.
Never mind the gender issues, I think it's funny that the entire basis for evaluating the quality of someone's code is how well-commented it is. How much documentation is the "right" amount to put in a piece of code varies *greatly* from situation to situation. A good programmer is one who knows when, where, and how much documentation is needed. Furthermore, it takes skill to make documentation useful as opposed to just redundant. The fact that this lady fails to mention these subtleties makes me think she... is not a good documenter.
On another note, everyone I've ever worked with has a different style. Some are terse, some are verbose; some are straightforward, some are clever. Everyone does it their own way. The difference between men and women is that I've known enough male programmers to be absolutely convinced that they're all different, and I've known few enough female programmers that I could see being tempted to generalize about their coding styles.
Agreed. I bought one for myself when I started working from home. I made a point of testing out all of the highly recommended office chairs that I could find and the Leap was a clear winner. The back support is fantastic, the arms are at least twice as comfortable as any other arms out there (important when you type and mouse all day), and the seat is flexible and padded in all the right places. This is close to a perfect chair.
The Aeron is cliche but it really is a great chair. It's my second choice after the Steelcase Leap.
One piece of advice based on personal experience: try each size before you buy one! In addition to the adjustments, Aerons come in three sizes-- A, B, and C. According to their sizing chart, I am size B, but in practice the C fits me like a glove. Big, big difference between sizes.
being exposed to high levels of microwaves can cook you, so I'm sure low-levels aren't entirely risk free the same thing could be said of "heat." just because you'd cook at 300 degrees F doesn't mean you're slow-roasting yourself to death at 72 degrees F. (in fact, microwaves are specifically used to create heat, not to make any kind of direct changes. they are just a convenient way to dump energy into the water in food.)
No kidding. Last time I was in Santa Fe, I got in an argument with one of the ladies behind the counter in a rock shop over this very issue. I pointed out that we are constantly bombarded by so many different kinds of electromagnetic radiation that it's impossible to single out wi-fi as a factor. I pointed out the low amount of power that those signals have compared to other pervasive forms of man-made radiation. I mentioned the number of studies that have been done and the lack of conclusive findings. None of it made any damn difference except that she got pretty cranky with me. She insisted that a bunch of her friends were allergic.
Your rocks will work, but only if presented with enough authority to sway those people. You need to find some wizened old woman from a Wise and Ancient People, like something South American or Native American, and get them to pitch the idea for you. If you can work a weekend intensive in there somehow, the deal is sealed.
i love the idea of someone trying to "take over" craigslist. craigslist is not so much a site as it is an idea. it's a glorified message board with categories and a "report this" button. if someone takes it over, they can do one of two things: (1) leave it exactly the way it is, in which case everyone wins, or (2) fuck it up somehow, probably with fees or ads, in which case someone starts a new craigslist elsewhere that has exactly the same feature set as the old one. the only thing that ebay would gain would be the rights to the well-known name "craigslist," but even that would become pretty worthless as people moved on to the new site.
Actual printing of semiconductor devices is still in the early research and implementation phase at various corporations and universities. Printable motors are the remaining practical hurdle. lol. i'm almost done building a space shuttle out of papier mache. the only parts i'm still working on are the engines and the heat shielding for re-entry. consider this a 0.1 version.
Don't shed any tears on my behalf. I bought an HD-DVD player right before the big WB announcement came out, and I can't say I regret my decision. I bought it because it was cheap enough to fit my budget and there was a pretty nice selection of movies on HD-DVD *already* that I wanted to watch in HD. Even if nothing ever gets released on HD-DVD again, I now have a nice upscaling DVD player, a small pile of HD-DVDs, and a medium sized pile of HD-DVD rentals on Netflix. It was worth $200. By the time I start chafing at not being able to watch the newest HD stuff, it'll be time to upgrade to a new player anyway, and Blu-ray will be cheap by then (or replaced by downloads).
Slashdot Mirror
Sphinx by itself is a terrible answer to this problem, unfortunately. The code is free, but good luck finding an appropriate model. Worse, you'll need to train a speaker-dependent model to get any usable results, and this is a VERY non-trivial task with Sphinx tools in the state that they are. I spent several years getting paid to adapt Sphinx for commercial purposes and while it's great for some things, I can say with confidence that it is not the tool you're looking for.
You know what works? Dragon. Hate to say it, but the commercial products here have a gigantic edge on the competition.
That said, I'd love to see someone come up with an open source speaker-dependent model training system that's friendly enough for app developers (not speech researchers) to roll into projects. I think this is a big open door for contribution to the community. Sphinx isn't the best thing going, but it's certainly usable, and if a real product came into being I'm sure all the speech wonks would start coming out of the woodwork to improve the algorithms.
Algorithm revised in light of real-world performance constraints! Read all about it!
Seriously, we just rewrote a tree (that runs in a high traffic serving environment) this month at work because it wasn't streamlined just right to take full advantage of the underlying architecture. No one will write a paper about it.
Also, hey kids, profile your code.
Even if generation 1 electric cars are a mess, SOMEONE has to make them viable enough and sexy enough to get the market moving in that direction. As soon as electric cars start becoming a significant chunk of the automobile market we'll see battery, motor, and material research go through the roof. This has already started, and I think it would be hard to argue that Tesla hasn't played a big role in swaying the general perception of electric cars from "slow ugly thing that hippies drive" to Serious Business.
Aptera is awesome in their own right, and you're right that their design pushes the envelope a lot further. Hopefully gen-2 mass market EVs will go more in that direction. I just don't think anyone should downplay the importance of Tesla (etc.) making the generation of cars that bridges the gap between what we drive now and what we'll drive once electric cars are pervasive.
Part of my password stays the same; part changes. The part that changes, I write down on a post-it (literally). The part that stays the same is memorized.
In practice, I have a sandwich, xxxxxYYYYzzzzz, where x and z are constant and Y changes to meet the needs. This is also how I customize for different applications, e.g. my slashdot password might be xxxx/.zzzzz, my bank password xxxx$$zzzz, etc.
It works. It's pretty safe and easy compared to the alternatives.
"FB can't possibly keep backups of every state of your profile"
Wrong, for two reasons: (1) as sites and traffic grow, old data becomes a smaller and smaller slice of the total storage pie, and (2) storage just keeps getting cheaper. Also... just how much space do you thing profile information takes up, anyway? I'm guessing photos and activity logs are the bulk of their storage costs.
Sure, .net would have been way better if it embraced Linux and welcomed re-implementation from the beginning. It would have been more widely adopted, there would be more open source VMs and IDEs, etc.
Why? Because people love free stuff. Good for the ecosystem doesn't necessarily mean good for Microsoft. No one wants to pay for a Windows license for each machine in their datacenter. How many Solaris licenses did Java sell for Sun?
The way I see it, .net was yet another decent attempt at forcing Windows lock-in (and keeping already-locked-in partners happy and productive).
Meanwhile, somewhere off the coast of England, a colossal killbot picks up the news headline from a radio transmission and pauses to reconsider its mission of vengeance.
the most shockingly wrong way to put that
yeah, cannons don't shoot bombs.
oil futures aren't exactly a junk stock. good luck making an impact on those prices with a spam email campaign. there's a reason they usually target little stuff you've never heard of.
i'd never heard the term "model selection," so thanks for pointing that out. it looks like there really is some good literature to read on the subject.
the process described by the model selection sites i skimmed still doesn't adress what i was getting at, though. "choosing a model from a set of potential models" is only conceivable when your set of potential models (and set of variables to potentially be modeled) is well bounded.
to put it another way, take the smartest model choosing algorithm you can find, hand it a pile of data, and say "what do you make of that, smart guy?" i'm willing to bet that the answer is going to be along the lines of "wtf?" unless there is some sort of context or metadata provided along with the data to give the algorithm a hint of what it's looking for. am i looking for covariance between scalar values among regularly organized groups? am i looking for white rabbits in the image data from a camera? is this ascii or ebcdic or 8-bit PCM data? you can argue that these questions are trivial, that no algorithm can be *that* general, but that is precisely my point: all known algorithms require significant narrowing down of the problem space by human hands before they can begin to produce useful output.
if you had an algorithm that took *truly* semantics-free data in one end and spit models of regularly occuring features out the other end, you'd be halfway to general AI.
instead of "42" it'll come up with "rule 34."
the difference is that brains create new theories and models to describe data, whereas this article specifically talks about avoiding the need to make new theories to describe data. we still have no AI that can create theories and models and semantics on its own. i agree that when that happens, we'll have something exciting and new, but it hasn't happened yet.
Never mind the gender issues, I think it's funny that the entire basis for evaluating the quality of someone's code is how well-commented it is. How much documentation is the "right" amount to put in a piece of code varies *greatly* from situation to situation. A good programmer is one who knows when, where, and how much documentation is needed. Furthermore, it takes skill to make documentation useful as opposed to just redundant. The fact that this lady fails to mention these subtleties makes me think she... is not a good documenter. On another note, everyone I've ever worked with has a different style. Some are terse, some are verbose; some are straightforward, some are clever. Everyone does it their own way. The difference between men and women is that I've known enough male programmers to be absolutely convinced that they're all different, and I've known few enough female programmers that I could see being tempted to generalize about their coding styles.
Agreed. I bought one for myself when I started working from home. I made a point of testing out all of the highly recommended office chairs that I could find and the Leap was a clear winner. The back support is fantastic, the arms are at least twice as comfortable as any other arms out there (important when you type and mouse all day), and the seat is flexible and padded in all the right places. This is close to a perfect chair.
The Aeron is cliche but it really is a great chair. It's my second choice after the Steelcase Leap. One piece of advice based on personal experience: try each size before you buy one! In addition to the adjustments, Aerons come in three sizes-- A, B, and C. According to their sizing chart, I am size B, but in practice the C fits me like a glove. Big, big difference between sizes.
No kidding. Last time I was in Santa Fe, I got in an argument with one of the ladies behind the counter in a rock shop over this very issue. I pointed out that we are constantly bombarded by so many different kinds of electromagnetic radiation that it's impossible to single out wi-fi as a factor. I pointed out the low amount of power that those signals have compared to other pervasive forms of man-made radiation. I mentioned the number of studies that have been done and the lack of conclusive findings. None of it made any damn difference except that she got pretty cranky with me. She insisted that a bunch of her friends were allergic. Your rocks will work, but only if presented with enough authority to sway those people. You need to find some wizened old woman from a Wise and Ancient People, like something South American or Native American, and get them to pitch the idea for you. If you can work a weekend intensive in there somehow, the deal is sealed.
i love the idea of someone trying to "take over" craigslist. craigslist is not so much a site as it is an idea. it's a glorified message board with categories and a "report this" button. if someone takes it over, they can do one of two things: (1) leave it exactly the way it is, in which case everyone wins, or (2) fuck it up somehow, probably with fees or ads, in which case someone starts a new craigslist elsewhere that has exactly the same feature set as the old one. the only thing that ebay would gain would be the rights to the well-known name "craigslist," but even that would become pretty worthless as people moved on to the new site.
Mr. Spacy Pants!
Don't shed any tears on my behalf. I bought an HD-DVD player right before the big WB announcement came out, and I can't say I regret my decision. I bought it because it was cheap enough to fit my budget and there was a pretty nice selection of movies on HD-DVD *already* that I wanted to watch in HD. Even if nothing ever gets released on HD-DVD again, I now have a nice upscaling DVD player, a small pile of HD-DVDs, and a medium sized pile of HD-DVD rentals on Netflix. It was worth $200. By the time I start chafing at not being able to watch the newest HD stuff, it'll be time to upgrade to a new player anyway, and Blu-ray will be cheap by then (or replaced by downloads).