Slashdot Mirror
Mac Version of NaturallySpeaking Launched
WirePosted writes "MacSpeech, the leading supplier of speech recognition software for the Mac, has canned its long-running iListen product and has launched a Mac version of Dragon NaturallySpeaking, the top-selling Windows speech recognition product. MacSpeech had made a licensing agreement with Dragon's developer, Nuance Communications. The new product is said to reach 99% accuracy after 5 minutes of training."
MacSpeech, the leading supplier of speech recognition software for the Mac, has canned its long-running iListen product and has launched a Mac version of Dragon NaturallySpeaking
Tell me more about has launched a Mac version of Dragon NaturallySpeaking.
The theory of relativity doesn't work right in Arkansas.
I tried Dragon a number of times, but it feels too much like talking to oneself. Training it is a chore too. 99% accuracy after 5 minutes is probably true, but I type much better than that. I suppose it will be great for people who either can't type properly or are lysdexic.
Excuse me, but please get off my Pennisetum Clandestinum, eh!
the whole intention of Dragon? For those people who *are* impaired in some way or another? I mean...I could never "speak" out a paper or something. I'd end up tearing my vocal cords out.
'Number-memorizing Chinese people.'-Anon
...welcome our new Dear aunt, Let's set so double the killer delete select all
...Mac users will have no trouble chatting with their computer for 5 minutes. Think of how accurate the system will be if the users got into a heated debate!
Obligatory blog plug: http://www.caseybanner.ca/
Will it recognize metrosexual accents?
I was about to say 13256278887989457651018865901401704640, but it appears this number is private property.
"I suppose it will be great for people who either can't type properly or are lysdexic."
Getting First Post! will be a lot easier.
The last time I tried using voice dictation was When I was running OS/2 Warp 4. Training took forever, and the experience of using it was nothing but an exercise in frustration, ending with me screaming at the bloody thing then seeing neat, yet random expletives on my screen. I later came across some budget software that required no training, yet worked surprisingly well compared to the $400 packages made by the big boys. That software really showed what voice diction should be like, if only it was developed further.
The training an accuracy seem like things that can be overcome, but I would really like to see a solution for things like punctuation and function keys, things that don't naturally come with speaking. Instead of having to say "delete that" or " delete" it would be nice to just have a button that I can hold down when saying things I want interpreted as commands.
"Computer... computer... hello computer?"
Am oosing it two type this comment. Didn't knead the fave mins train ming though...
Acid House saves Souls
i know the answer. No it doesn't.
I own a copy of dragon 9 but having to reboot into windows to use it makes it too much of a hassle. Wine doesn't seem to handle it either.
It actually works quite well, although mileage may vary depending on the sound quality you get from your microphone, soundcard setup.
I'll have to play with Dragon at some point; I just haven't gotten around to it yet. Aside from accuracy errors, the primary issue that bothers me about speech recognition solutions I've tried is the general lack of being able to recognize speech that seems natural to humans but isn't what the system is expecting as input.
This is especially true with over-the-telephone solutions. For example, I am with Rogers Wireless carrier here in Canada, and their automated customer service system prompts you for your phone number. My last 4 digits are 2125, and it is very natural to say "twenty-one, twenty-five" when giving the number to a human being. The speech system, unfortunately, is only sophisticated enough to understand one-digit-at-a-time mode, so you have to suffer through saying "two one two five". Which isn't truly a big deal, but it's frustrating having to learn each system's unique quirks and limits. I suppose the same can be said of any technology.
Oral dictation (as opposed to fixation) is frustrating at best. Punctuation is a critical item that I can't stand dealing with. Trying to get the goddamn software to insert commas and semi-colons can be difficult enough, let alone wanting to actually insert the word "comma" into a paragraph. Then there's trying to spell out acronyms (aka "aka"), or inserting the contents between and including those parentheses. Until dictation of a document can be done with truly minimal correction and post-editing, and can be spoken at a very comfortable pace, I will stick to a keyboard.
Of course, the most entertaining aspect of watching someone else play with speech recognition is the inevitable habit of sounding completely unnatural while speaking. The monotone voice and sounding like a robot are bad enough, let alone those who think that shouting or talking ree... aaa... llll... lllyy... sloowwwww.... llly is going to help. The funniest I've seen was a woman who seemed to think that talking in cutesy baby-talk would win the system over to her side. :)
I just want a system that responds to commands via a programmable keyword. Only when speech recognition is Star Treky enough to respond to its name will I be happy. My computer will be named Minion.
iIt iworks iso iwonderfully iand iintegrates iwell iinto ithe iother iiproducts.
But 99 out of 100 words correct still makes for a pretty lousy experience if you're trying to do anything serious.
Personally, I think so much when I'm writing that typing is quite fast enough. Of course, I know not everyone is so fortunate.
expandfairuse.org
Yawn. We had it about a decade ago.
> The new product is said to reach 99% accuracy after 5 minutes of training.
According to MacSpeech, I suppose?
I'll bet what was said was something 99% different to what MacSpeech thought.
Max.
I was at the Apple Dev conference in 1999 (or so) when the CEO of Dragon got up during Steve's keynote and announced that they were going to develop a Mac version of Dragon.
Almost 10 years later - and it's finally here!
Or at least a follow up announcement is here.
I was a bit put off by their pricing scheme. It's $50 off the normal price (something like $200) if you buy it at macworld. The only problem is that it's a pre-order, so you can't try before you buy. Also, nobody has reviewed the software, since it doesn't exist yet, so if it turns out to be a stinker you're out $150. And if you don't like the product, their tech support will try and "walk you through" your problem to make it go away. They explicitly said "no refunds". No, thanks.
Quid festinatio swallonis est aetherfuga inonusti?
Africus aut Europaeus?
- *Simple* speech recognition. I want it to react to a keyword ("Computer", or "House", or similar sci-fi-ey) and then a few simple commands. Sphinx-2 seems ideal, but I'd need good dictionary files.
- Ubiquitous microphones (preferably exclusively usable by the speech recognition engine. Setting proper
/dev permissions will help). Probably the most difficult/expensive to get right; it needs to work in noisy environments.
- Machine controllable electronics, sufficiently protected so that . Where those 433MHz remote switches come in I guess. Needs to be code protected, for obvious reasons.
- Scripts to tie all this together.
Has anyone done this properly/successfully/usefully?http://ars.userfriendly.org/cartoons/?id=20010322/
Surely you mean...
'Illuminate'
'Deluminate'
MacSpeech is scum that has been selling absolute shit for years.
Their iListen product was absolute unusable garbage, but that didn't stop them from marketing it as the Mac's equivalent of Dragon. Complete with "30 day money back guarantee" that meant you could get your money back only if you tried it full time for 30 days, and you convinced them that you had tried it full time for 30 days, and you had bought the special (shitty) microphone the software "requires" (I think they actually just had a marketting agreement with the manufacturer, because this TELEX headset was very low quality AND expensive), and they decided to give you a return authorization number, and the moon was in the house of Jupiter, etc.
Basically, nobody got their money back. Because they're liars and thieves who are used to selling garbage.
Truly unfortunate that this slime got the contract.
Back in the late 90's using only Applescript and the Apple built in speech recognition I was able to voice automate my music library. I don't remember all the details, but I could start and stop the music and select what artist I wanted to hear. It was pretty neat being able to say "Computer, play Nirvana" and getting my music all from the comfort of my bed.
We hope your rules and wisdom choke you / Now we are one in everlasting peace
Because I helped a illiterate person write a letter in jail and spelled everything out for him verbally... then I thought it was like he led me right through a training program for a speech recognition program.
?
The writers must have been using that software when they wrote this song!
Have you heard of VMWare ?
Be careful what you use as the trigger, or else you won't be able to use the words "House" or "Computer" in any conversation while at home without the house thinking you are trying to command it, and starting the dishwasher or something. I suppose you could always name your house something sci-fi-ish, or fantasy-ish that would never come up in conversations, like "Malthikar." For extra points, establish some sort of visual avatar piped to your TV or something so you can see him while you talk to him.
As for implementation, Mac OS X comes with some sample code for Dictionary based untrained speech recognition. Should do exactly what you want. Since you can give a list of all possible words (the various valid commands) it works better then free-form recognition for general text input. And, you don't have to train it, so anybody who knows the right things to say could work your house. That just leaves having your app do the commands once they are recognized. I'm completely unfamiliar with that end of things, but I know there are home automation doodads which presumably shouldn't be that hard to access from a program.
I've worked with Nuance's server product in the Dragon NaturallySpeaking line as a developer. Their API is confusing, their speech recognition SUCKS, and their software bugs out in bizarre ways. It's also slow as a dog, and advanced functionality (like recognizing from wav files, as opposed to from a live audio stream) is so poorly implemented as to seem bolted on.
And the worst part? Nuance has a virtual monopoly in realistically priced (read: "in a budget that a normal small-to-medium-sized business can afford") general-purpose speech recognition systems. If I recall correctly, they bought out Lernout and Hauspie's speech recognition products and IBM's old consumer-level speech-recognition stuff. So you can't take your business elsewhere; there is no "elsewhere".
I loathe those guys.
With spending like this, exactly what are "conservatives" conserving?
I had about a 98% accuracy rating with the included microphone and no sound card.
You know, there is a difference between trolling and pointing out the flaws in your reasoning. Just saying.
should cover most needs ;-)
Five minutes training for most people, but not everyone. My boss uses Dragon NaturallySpeaking, and it took him nearly two weeks to complete the five-minute training due to some complications.
Namely, he's blind. He cannot read the training phrases off the screen, because he can't see them. Instead he had to have a screen reader (JAWS in this case) read the phrases aloud to him so that he can repeat them back. But of course, Dragon was not expecting to hear audio input from anything other than the user, so that confused things. There were problems even using a headset. And since he can't actually use the program at all without having the screen reader running, it was pretty awful trying to get the training done. I'm not even sure how he finally managed to do it - I suspect he probably got a sighted friend to help. Thankfully the training files can be copied from one computer to another so you don't need to retrain it on each different installation.
Once the training was finally finished, it worked well. He has poor fine motor control as a result of leukemia treatments - he can type, but only slowly and with a high error rate. His speech is slightly slurred as well, which reduces the accuracy of the transcription. Even so, the Dragon transcriptions are definitely better than manual typing. It's helped him a lot.
I just wish that the Dragon programmers would come up with a more easily accessible training routine. There aren't a whole lot of users with the same disabilities as my boss, but for the few like him having good, well-trained dictation software is vital. With it, he can control his computer reasonably well, if rather more slowly than a sighted person with normal motor control. Without it, using the computer is basically impractical. When he can't use Dragon, sending a single rather short email can take upwards of an hour.
A lot of the comments here say something along the lines of "I tried it years ago and it was rubbish" - yeah well things have moved on!
It's like saying all cars still have wood framing and carburettors like they used to be - a contemporary vehicle is steel and has fuel injection - oh and an iPod dock...
Dragon 9 whilst still not perfect is really very good - and MacSpeech will build their product on that technology.
Oh and by the way the Google speech recognition is by Nuance too and ViaVoice whilst distributed by Nuance is an old seperate IBM product....
This software's history includes jail terms. Speech recognition has gotten an extremely bad reputation for being worthless garbage, maybe because it is worthless garbage.
Even a 0.5 percent recognition failure rate is enough to make speech recognition software worse than worthless. The reason is that speech recognition software never makes a spelling mistake. Instead, the mistakes are often extremely difficult to recognize, and sometimes change the meaning in subtle ways. That's partly because when the software is confused it tries to select something that is grammatically plausible.
The result is that it has become difficult to sell speech recognition software. A high enough percentage of people in the U.S. culture know that it isn't actually useful. The orginal owners of Dragon NaturallySpeaking sold the product to a company that sold it to the company that became Nuance, maybe because they felt the product was damaging the credibility of their trademarks.
Here is a quote from the ComputerWorld story linked in the earlier Slashdot story, Is Speech Recognition Finally 'Good Enough'?:
"In 1993 two executives from Kurzweill Applied Intelligence (which pioneered SR for the medical market) went to prison for faking sales. That firm was sold in 1997 to a Belgium SR firm, Lernout and Hauspie (L&H), which was reporting phenomenal sales growth at the time. Dragon Systems, which originated DNS that year, was reporting only anemic growth, and L&H had no trouble acquiring Dragon Systems in early 2000 in a stock deal. Within a year a series of accounting frauds came to light and L&H collapsed into bankruptcy. Its SR technology was sold in late 2001 to ScanSoft Inc., which kept the DNS line going. (It was then at Version 6.0.) ScanSoft later acquired Nuance and adopted its name.
"Thereafter, "It was with the launch of Version 8.0 (in November 2004) that the market became reinvigorated and took off," said Chris Strammiello, director of product management at Nuance. "We crossed an invisible line with Version 8.0, where the software actually delivered on its promises and offered real utility for the users. Sales have been growing at a rate of 30% yearly since then, except that we expect it to do better than 30% this year."
Read that again: "... the software actually delivered on its promises and offered real utility..." I called Nuance and was told that version 8 did not have a new recognition engine, but only had improvements in the user interface. A friend who owns and tested version 8 told me he could see no difference in accuracy between that and version 7.
So, in my opinion, Nuance has done common deceitful things that are called "Marketing":
1) Bring out new versions. Previously, when there has been a "new version" of Dragon NaturallySpeaking, I call Nuance technical support and ask if there is a new recognition engine. I didn't call for version 9, but for the last two versions they have said no. So, nothing is changed; the software is still worse than useless to me, in spite of the fact that they advertise that the software is now more accurate.
How is it possible that the software is more accurate, if the recognition engine did not change? Maybe it isn't true. Or maybe the company improved the guesses the software makes when the software really has no clue what the user said. As I mentioned, those guesses have become so sophisticated that you can become confused about what you actually said, and you have to spend time re-creating your ideas. If you are saying simple things about a simple subject, this is not as much of problem as when you are writing about contract negotiations, for example.
In the words of a Slashdot reader: "The opinions expressed here may be those of my speech recognition so
I don't think dictation's the solution. If you're discelyc what you really need is a spielchucker.
And what about about people who speak dyslexically? Yes, Dubya, as it happens I am looking at you.
Confucius say, "Find worm in apple - bad. Find half a worm - worse."
Seriously though, is it just me or is speech recognition support still sadly lacking under all current distros?
Based on the fact there are no leading edge projects out there. I mean, apart from IBM's ViaVoice a few years back (and now no more), and the CMU Sphinx project http://cmusphinx.sourceforge.net/html/cmusphinx.php is there any other Linux/FOSS solution?
In his case, that might be OK.
But for the rest of us- we choose to use Linux because we want to use Linux. For most Linux users, it doesn't make much sense to buy and install MS-Windows and Dragon to use in the free/open Virtualbox or the proprietary/closed VMware. With such a model, you cannot use the speech recognition in the Linux applications.
Understanding 99% of what I say correctly after 5 minutes is a lot better than the developers do...
This was first announced eleven years ago. It's about time. Maybe Pogue will stop using Windows now?
And it's amazing. I find that it's much more natural and fluid for language to go from thought to speech than from thought to typing. Also, the accuracy is better than typing, (including spelling) and it comes with a headset that is more than adequate. Give it a try folks, and forget about carpal tunnel forever...
This is fantastic news for those who need extra accessibility features.
It may be fine for you or me to hit any key, but there are many other folks with various disabilities for whom such a task is not an easy one. So it may make more sense for them to use their voice and move on.
If any of us were to lose fingers or hands in an accident, I bet we'd all be using something like Dragon to continue our work, rather than try to become a tap dancer.
And let's not forget about accessibility in the workplace. This is great news for Mac shops, as now there is one less reason for having to support a rogue Windows machine...
If you don't know what AltaVista is (was), get off my lawn.
It's fine to port this to the Mac. Fine. Good. Whoopie.
But they are so DROPPING THE BALL. They have the best voice-rec platform. (You can think it's not good enough, but it's still the best.) What they need is to port it to Linux. Duh! Wake UP!
No, I'm not just saying the usual "Does it run on Linux?" bit. Linux is the now (and coming even more) obvious OS for small devices. When you want to talk to ANY device in your home or car, or your cell phone or PDA, you'll be talking to LINUX. THAT'S where we need a great voice-rec system. We need it ported to Linux and opened for an API. This will catapult this annoying desktop app into a present on almost everything type software device in a matter of a couple of years -- as low power devices provide enough umph to do what the heavy machines of a few years ago do.
My wife needed voice dictation software a year or two ago. She had been a Linux user. I gave her my PowerBook and bought iListen for her. It was terrible. And it was a resource hog. It used the Philips engine and, even with extensive training, was the pits. We even tried several high-quality mics to no avail.
She went from my G4/1.5GHz/1.25GB RAM PowerBook running iListen to Dragon NaturallySpeaking 8 on an IBM ThinkPad T23. (P3 1GHz, 768MB RAM, WinXP.) The difference was night and day. Not only did Dragon run much faster on the lowly P3, but the quality of speech recognition was _much_ better. As a result of this, she's now back to being a Windows user with Dragon.
At least it looks like our iListen purchase won't be a complete waste, as we can use it to upgrade to NaturallySpeaking for Mac. I'm glad that MacSpeech has killed iListen. It needed it. It was an embarrassment compared to Dragon.
Speech recognition has been a big hole in the Mac's software line-up. It looks like that is finally coming to an end. Now if only someone would release something that works for Linux.* I know that we'd have paid $200 for something approaching Dragon 8's capabilities.
----
*Yes, I know about IBM ViaVoice. Good luck getting that to work on any recent distribution. I also know about Sphinx. Unfortunately, it seems to be a perpetual research tool rather than an end-user program.
Dragon had a Mac product once before - Dragon Power Secretary. It was tied to specific apps. Didn't get much updating or new versions after the initial release and died an agonizing death.
"Enjoy what you're doing! If it becomes drudgery, you're doing it wrong!" - Jim Butterfield
Here's hoping they support Linux next.
I have used their speech synthesis products and they're quite impressive. I used one of the voices to dictate a textbook into an MP3 file so that I could then do a book-on-tape type thing to play my textbook in my car. The pronunciation was generally pretty good. I had to define the pronunciation of a few words here and there (it had problems with some of the less common geek words, like "macromolecular"). But after giving it the proper pronunciations, it was quite excellent. The voice sounded natural a good portion of the time.
Eye lie kit mice elf.
About that 99% ... in honor of his Steveness, who tested the software while writing a recent keynote, one out of every 100 words is "Boom." But in true Mac fashion the options panels are severely minimalist, so the Boom feature cannot be disabled.
I am looking for exactly this, except for Windows XP. I don't need speech recognition for anything but playing music and it seems like this should be completely possible.
Do you happen to know of any (preferably free) software or plugins for existing media players that will do this? I would like the basic functions such as play songname, play artist, play genre, pause, next and previous.
Macs have been lagging behind for years now. With the advent of Windows Vista, complete with a 16kHz recognizer and "easy-to-use" integration into practically every section of the OS, MAC had no choice when it comes to "Keeping up with the Jones" ;-)
Your Audio Content, Live Streams, and more... To every phone: via Shout-Outs, or On-Demand www.PhonePortals.com -
It should be noted that the 1% inaccuracy includes uncommon words that people never use like 'damn' 'this' 'stupid' 'software' 'that's' 'not' 'what' 'I' 'said' 'delete' 'no' 'don't' 'write' 'that'
I have nothing compelling to say
the built-in s.r. works great for this...lifehacker.com has a good how2 on creating app.specific voice cmds.
the other issue is microphones: only built-in or hardwired work...i tried a bluetooth headset, and while i could successfully pair & tx/rx, the s.r. setup wouldn't calibrate:-(min.level was 1/2 scale)-:
then i read that b.t. is too noisy...
Reading the comments I'm see a bunch of tabs[1] with no clue about being disabled, the speech recognition market, the history of the product, and how nuance is probably hampered by the management attitude towards money and the history of the code base.
for someone who's been disabled (temporarily or permanently) speech recognition means the difference between making a living and being able to support oneself, a mortgage, family etc. and sitting around on your ass in section 8 housing on Social Security disability. Pain from RSI once made it extremely difficult to feed myself. When you've experienced that level of pain, disability and the associated despair, you get the attitude that anything that gives a disabled person independence and an ability to make a living should be encouraged with all possible resources.
Listening to someone dictating using speech recognition will drive you mad. You would have the same problem with a blind person listening to text-to-speech. But that's not the fault of speech recognition or text-to-speech. That's the fault of management not providing the disabled person with an acoustically isolated environment (i.e. reasonable accommodat.
Desktop speech recognition is a monopoly because it's extremely expensive and difficult to develop speech recognition and there is not a large market. the market consists of lawyers, doctors, and the disabled. There is not enough money to support two companies (or more) to develop desktop speech recognition applications.
NaturallySpeaking is very buggy. There are bugs that cause people problems that were first seen in NaturallySpeaking 5. These are not hidden or hard-to-find bugs. They don't affect nuances ability to sell NaturallySpeaking. There's no reason for them to fix them except for the fact that they interfere with the use of many programs by the disabled. If you are just doing dictation into Microsoft Word or DragonPad, you'll never notice. If you try to dictate into Thunderbird, Firefox, Open office,... you're screwed. For example, I cannot dictate directly into Firefox for this comment, I need to use a workaround for dictation and then paste the result into the text box. The reason why this problem exists is because nuance management has the reputation of not making any change or feature unless you can make a business case and show them they will get revenue from that change. This is not such a bad model because it can keep nuance profitable and product available to people who truly need it (i.e. the disabled). The downside is that it doesn't leave room for changes necessary for the disabled.
I've heard from people working inside dragon that part of the problem also is the code base. It was written by a bunch of Ph.D.'s who are really really good at speech recognition but are not so good at writing code. Also in the last few years, there has the huge turnover and people working on the code as NaturallySpeaking was sold first to L&H and then to nuance. That kind of change alone will wreak havoc on the code base as knowledge is lost and never really acquired by the new people. by the way, I have talked with some people from nuance, and they are basically good people. They understand the needs of the handicapped but they are constrained in what they can do for us because of budget and resources.
When people talk about alternatives with open source speech recognition, only a tab would think they would work for the disabled. Their recognition speed is significantly slower, vocabulary size is smaller, and they are really more projects to keep grad students busy than be anything useful in the real world.
The last problem with speech recognition sits in your lap if you are a manager of a software product or a developer. As far as I can tell, the number of applications that are speech recognition friendly is vanishingly small. It seems to me that software developers go out of their way to make software handicap hostile. It starts with the multiplatform GUI toolkits that do not
The top selling speech recognition software is the one that comes bundled with every copy of Windows Vista.
Now Mac users can enjoy the merciless flow of spam that PC users enjoy. "LAST CHANCE!! Nuance blah blah blah". AAAUGH!
Citation and relevance, please.
In my job, I teach clients how to use this software everyday.
The 99% accuracy is after the initial training. Then comes the tutorials which further enhances recognition and use, which makes even more accurate. Dragon is invaluable for those would cannot use a computer any other way.
Accuracy does increase with time and use.
CDE open sourced! https://sourceforge.net/projects/cdesktopenv/
SO, now that Dragon has been ported to Mac, it is only a little more time till it is ported to Linux.
I have searched for a good bit of software in Linux to do voice recognition, but nothing is a) easy to use, and b) (in my opinion) not ready for market. This has left me with using Microsquat's OS and Dragon. IBM also has ViaVoice, but it isn't for Linux yet, either...
Still waiting
--E--
I do know that David Pogue uses DNS for all of his writing (NYTimes, books, etc.). He writes about Dragon often, and how he previously used to carry a Windows laptop essentially just for writing.
Many Mac folks, myself included, have installed windows via Fusion or Parallels so that we can run DNS alongside OS X. I have got it working reasonably well, and have been doing all my writing and email via speech for about 6 months. There are still some frustrations, but in general it works great and I'm happy to have it.
The big question I have is whether this version will be better than using DNS via a virtual Windows machine. Unless the implementation is horrible, I'd expect so. Well, reasonably I would expect it to be crappy until a few versions in, but I'll be watching closely for reviews.
I work as an Assistive Technology Trainer, working with HE clients who have a variety of disabilities, from severe physical difficulties to learning difficulties such as Dyslexia.
I can understand the negative comments about voice recognition. Generally, they are based on poor experiences with Dragon (or god forbid ViaVoice, or iListen) as it was a few years ago. It is only since version 7 that recognition accuracy has increased to a useable level.
At present, we use DNS9, which, let me emphasise now, is BLOODY EXCELLENT.
Within 10 minutes of training, which involves reading a segment from a text, accuracy is between 85-95%. Further refinement comes when using the program after the training. This is a simple process, where the user can choose alternates to misrecognised phrases/words ("ice cream" for "i scream") for example, or misrecognised words can be selected and spelt by speaking the letters.
Once you have used the program for a week, it is 99% perfect.
Speech recognition depends entirely on the QUALITY of the headset you use, the POSITIONING of said headset, and the AVAILABLE MEMORY of the machine. In my experience of use since version 7, it these 3 factors alone that cause untold difficulties. Dragon used to come with the crappiest headset known to man, which was supplied with the product, in the box. Standard analogue jacks, and rubbish microphone. What you need is a USB Plantronics headset, postioned correctly, and at least 2GB of RAM. Recognition will be perfect, i promise you.
My clients find Dragon enormously useful. For example, a student I was training was severely disabled, wheelchair bound, and had only the power of speech. With Dragon Naturally Speaking, she could COMPLETELY control her PC, from dictating documents/email, using the internet, opening folders etc. Everything a user using a mouse can do. Dragon allowed her to open a window on the world.
In addition, Dragon is one of the most intuitive, and straightforward pieces of software ive ever used. It does exactly what it says on the tin, and within 10 minutes you are up and going and it works perfectly. Unless you have VISTA of course, but thats another story!
Cheers
Tim Symons AKA Anonymous Coward
I use voice recognition in dictating radiology reports all the time. It actually works pretty well. It is faster than typing, particularly with voice macros. Maybe because we have a limited highly specialized vocabulary...
Citation and relevance, please.
If you're going to throw out this tired old "i can't be bothered to google" shit, at least you can be bothered to read past the first sentence. Like to the part just after that where he links to the computer world story and quotes the part about the executives going to prison.
Relevance? He explained how after Nuance took over, they were selling the exact same recognition engine with a prettier UI each new version and claiming that each new version had a better recognition engine, giving anecdotal evidence that claims that two versions produced the same results. Is it different with the recent versions? Who knows, as they say in the stock market, past performance does not indicate future returns.
I have to admit though, that it would have been much more interesting if he'd found a case of someone going to prison because the software mangled their words into something else.
Focusing on using speech recognition software for dictation has always seemed to me an aggravatingly limiting use for such technology. I want to use voice recognition at the command line, not as a substitute for typing a document! In that setting, 99% accuracy will be sufficient.
Even the voice recognition that was included with MS Office was accurate enough to recognize the far smaller dictionary of command terms I would need to use voice recognition as an OS interface, with accurate letter recognition and auto-complete features for navigating the file structure.
Piping terms into the shell may now be a possibility.
My experience with using speech recognition for commands is, well, not good. You really need *better* accuracy than you do for transcription. But if that's what you want, it'd be trivial to implement. You could possibly do it with Applescript and the existing tools.
Personally, I want it for transcription of recorded speech, not real-time transcription.
Sometimes software is just a useful solution to a need.
When I was in college, I had to have unscheduled surgery on my right elbow the week before the final paper was due for my Anthropology class. The outline for the paper was complete; I just needed to type it out. Being in a cast after surgery made typing impossible but with one hand. I purchased and used DNS to finish my paper. The software was easy to use and errors were minimal, even for an Anthro paper with lots of jargon.
I got an A on the paper.
My comments here are my own; I do not speak for my employer.
I actually bought Dragon naturally speaking 9 about a week ago, and have been using it in the evenings at home since. Of course I was sceptical about how accurate the software would be but I thought it was worth a gamble for the £40 asking price for the Standard Edition. I have since dictated only around 3000 words using the software and already I can dictate faster than I can type (50 words per minute typing) and the accuracy seems to be around 97% on sample articles. What also encourages me is that the accuracy has already increased and seems to continue to be increasing, so on that basis and having only dictated a short document's worth of words, I expect this software be very useful to me. I don't have any disabilities, but thought dictating would be more comfortable than typing for extended periods. However I can already imagine that for people with disabilities, RSI or sight problems this software would be extremely useful to them. While it isn't perfect I would argue that it is of production quality.
If you have a Mac with a microphone (which should just about cover every consumer Mac made within the last decade), turn on speech recognition and say the words "Tell me a joke".
Have fun.
Accuracy may improve without a new engine in various ways. The software may have a better dictionary, the engine may have been trained on more data, and the program may include better noise filters.
Also, speaking like a robot will actually hurt the performance. Software like this has been trained on real-world audio data. In fact, speaking overly clear is a mistake people often make when speaking to speech recognition software. The software will handle casual, daily speech better than overly formal or "robot" accents.
Let's see,
I've tried them all - and spent dozens of hours and hundreds of dollars for "custom dictionaries" - as a lawyer I dictate in a language that is almost, but not quite totally, unlike English.
I've had excellent "luck" with iListen and almost bought their "last" upgrade a few weeks ago. I decided not to because I couldn't get the helpdesk to tell me if my custom dictionaries and profile would transfer to the latest release.
I know why they didn't answer - and I'm not alone - the support mailing list posted this response today:
I can give you some basic information right now. MacSpeech Dictate is
a brand-new product written from the ground up around the same engine
that is in Dragon NaturallySpeaking. The new product is Intel only.
There was just no way for us to bring out a product based on the
Dragon engine and make it anything but Intel only. Besides being an
Intel only product you must be running MacOS 10.4.11 or higher. In
other words the last version of Tiger or a version of Leopard. Minimum
processor speed and memory requirements have yet to be determined as
we are still in development.
--
Technical Support Manager
MacSpeech, Inc.
Check out our online Helpdesk at:
http://www.macspeech.com/support/
Guess who is throwing in the towel? I've had it with the whole lot. Wake me in 10 years when somebody actually has a product that works. (And, don't blather at me - these bozos still don't know the final memory / processor configuration their new program will run on!) I suppose a Mac Pro 8 Core running at 3.2 gig and having 32 gig of RAM might suffice for that 99% accuracy - BUT that Mac Pro with a display and standard hard drive / standard graphics card + warranty would only cost 14,766.95
... the guys who've promised german language support for iListen nearly ten years ago.
They just didn't deliver (as far as I know, as I quit paying attention to them some years ago).
A usable solution in the field of speech recognition would still be a very important feature (something like a "killer feature") for any desktop OS, be it Mac OS, Windows, or Linux.
It is indeed funny how people talk when they use ASR. I work on speech recognition in grad school and we do usability experiments and people need to be explicitly reminded not to talk like a robot. the performance is best when talking normally, but because people attribute human-like qualities to the recogniser, they talk to it like it was a robot child that didn't understand, which makes the performance a lot worse. But it should be stressed that it isn't the fault of the recognition that people talk like robots, it's just what people do naturally when they think there's a lack of understanding. --why procrastinate today when you can do it tomorrow?
How is it possible that the software is more accurate, if the recognition engine did not change? Maybe it isn't true. Or maybe the company improved the guesses the software makes when the software really has no clue what the user said. As I mentioned, those guesses have become so sophisticated that you can become confused about what you actually said, and you have to spend time re-creating your ideas. If you are saying simple things about a simple subject, this is not as much of problem as when you are writing about contract negotiations, for example. I'm in the biz, and I'm not about to start defending Nuance, but this comment is a bit unfair. The underlying recognition engine doesn't have to change for accuracy to increase. The way that these systems are trained is to take a large amount of speech data and build models of the way that people speak. These models are plugged into the engine, which one can think of as like a big search engine, but instead of searching over a document collection, one searches over possible word sequences.
Two possible ways that improvements can occur without changing the engine: if I train baseline models with 10x or 100x the original data (collected from a large number of users, with a variety of accents, vocal tract lengths, etc), my statistical models of what a
That said, yes, there has been a long problem with over-hyping of products. Speech recognition has helped for many different domains (and yes, I can gripe about lost-luggage systems like everyone else), but clearly we as a field have a long way to go. Language is complicated, and yet people (usually) have no trouble using it, which is what makes it such an interesting thing to study and work with.
I'm a radiologist who uses a Nuance product for several hours a day, every day, and my experience has been overwhelmingly positive. Whereas I used to waste a great deal of time editing and correcting mistakes by human transcriptionists, I only occasionally have to manually correct the Nuance transcriptions. Our throughput and efficiency have increased considerably since we started with the product, and there is absolutely no way that I'd ever return to the previous system. The adoption of speech recognition has been the biggest advance in my field since digital imaging, IMO. Oh, and "when the software is confused it tries to select something that is grammatically plausible"? I don't think so - the software has no concept of grammar.
After reading David Pogue's review of Dragon Naturally Speaking in the New York Times a few years ago, I had been considering that app as the only reason I personally have that I would ever want to buy a Windows machine. Now that the Macs can simulate a PC, I finally went ahead and installed Windows on a partition on my Mac recently and bought the Windows version of DNS. It works pretty good for me. I use it to transcribe telephone interviews I do for my work (I can capture Skype audio) -- playing the conversation back to myself off my iBook and repeating both sides into DNS on the Mini. It has problems -- like, my little Mac Mini with the factory 1 gig of RAM is a little slow; and it's tricky. Like, I have to wear earplug phones *and* a headset mike -- I won't go into it all, but so far it is definitely awkward. However, it does work. It's faster than using my fingers, and although can be frustrating with the mistakes, it is close enough most of the time.
I've been working on getting the Mini to run the Mac OS and Windows at the same time using Parallels, so I can play my recordings and transcribe on the same computer. No dice so far -- the Mini is too slow, there are conflicts with the USB mike, and other problems; and I have other things to do than mess around with it.
So it's good for me that DNS is now available for the Mac. Although it does gets my ass that I could have waited six weeks and bought DNS native for the Mac instead of going through all that Boot Camp and Parallels crap. Figures. But yeah, let me have it. It's a productivity tool, for this workin dog at least.
I can't wait until I can drive my robot by walking on my treadmill with needles in my skull. Now that -- that will be progress. I will totally have the power.
I recently watched a demonstration of Utter Command which is an extension of Dragon NaturallySpeaking and it was pretty powerful.
"... uses a Nuance product..."
Which Nuance product are you using? Is it a special medical version of Dragon NaturallySpeaking? Those are much more expensive, and I've never tried them. They also use special dictionaries provided by Nuance.
I've heard about success with that use. Partly the success seems to be due to the fact that there is never confusion about what you said, so that mistakes are easily corrected. Another reason is that technical words are much more easily recognized.
Speech recognition software checks the grammar to see if the use of a word is plausible.
Does anyone have any experience with speech recognition software in other languages? It seems to me that languages such as English with it's many exceptions to pronunciation rules and it's grammar would make it a more difficult problem to solve than a language such as Japanese which has a smaller subset of sounds and more parsing friendly grammatical particles (post-positions as opposed to prepositions) that mark the word's part of speech. Does the problem become easier or harder based on the language?
Yowza, did those sleazy Belgians ever take him for a ride!
That is such nonsense. I suppose that if one is an excellent typist, the 99% accuracy rate is not great. But as a person who never learned to type, and has dictated to secretaries (some of whom generated work with less than 99% accuracy), I find the latest version (9) of Dragon Naturally Speaking to be outstanding. I do not find it frustrating at all. Frustrating is using two or three fingers to type more than a sentence or two.
The accuracy issue is overblown anyway. Responsible people proofread and revise what they produce, especially where documents are important.
The robotic talking habit is more ridiculous tripe. Never, because you don't have to talk like a robot to use the software.
You can actually generate work faster that people who type. And some typists get repetitive strain injuries.
Voice recognition may not be for everyone, but I think it is fantastic.
Well I've been told that iListen is great. I wonder why would they pull it?
It's true, voice recognition may be useful to people who don't know how to type.
"The accuracy issue is overblown anyway. Responsible people proofread and revise what they produce, especially where documents are important."
It depends on the material. In some kinds of dictation, voice recognition software makes plausible mistakes that may be missed by a proofreader, but make a big difference in meaning.
You are very unlikely to be aware of changes in your own accent.