13 replies to this topic

[bwelford]

I've now been using Dragon Naturally Speaking (DNS) voice technology to dictate blog posts for some months. The system has slowly become more used to my British accent and now performs reasonably well. However there are still some minor irritations and I wondered whether others had found solutions to them.

One of them is that you can train the system and teach it words that it should recognize. The only word that creates a problem is the word of. Actually I'm dictating this using DNS and as usual it displayed off. Somehow I cannot get this one right. I have trained it on the single word 'of' but it seemed to have no effect.

Another thing is that occasionally it will guess a phrase I have used and get it completely wrong. If you then mark this with your mouse, it will show you what other alternatives it might have used. So often I find that item 2 or 3 is exactly what I thought I was saying and yet the system produced something entirely different as a first guess. Does anyone have an answer for that?

I am using a fairly cheap microphone and I have wondered whether upgrading might improve the total performance. Again I would be interested in others' reactions to that.

Overall I'm a pretty happy camper. This whole Forum post is something I am dictating as I think. .. and mostly DNS got it right first time.

Is anyone else using this or other voice technology and how do they rate it?

[EGOL]

I have used DNS for a few years. As a professor I used it to get text versions of my lectures. Also, I used it to dictate while traveling. I had one of their voice to text recorders and carried it to dictate items for my "to do list". It was able to understand my Appalachian accent and after two or three days of editing it made very few errors (as long as I avoided some contractions that a person shouldn't use in formal writing).

I don't talk to my blog. My posts are too short. But, perhaps I should use it for email... hmmm...

Edited by EGOL, 06 May 2008 - 03:54 PM.

[Ron Carnell]

I'm curious, Barry, which version and level you're using?

It's been more than a decade since I played with voice recognition software, but your experiences, Barry, prompted me to consider it again. I just spent about an hour reading reviews (mostly at amazon.com) and deciphering the differences between Nuance's various software levels. It's probably not what I want, but I thought I'd pass on a few links I discovered while searching.

Here are the product reviews for Version 9 Standard and Version 9 Preferred. I think there's some really good comments in the User Reviews section, Barry, about the impact a sound card and microphone have on the efficacy of the software. One of the reviewer also mentions that Nuance has a forum for their products, though at the same time complains that it's not very active. Might be worth checking out, though?

I also found three books carried by Amazon.

Dragon NaturallySpeaking for Dummies

The Dragon: NaturallySpeaking Guide ...

Dragon Naturally Speaking for the Office Professional ...

All three include Amazon's "Look Inside" feature, so you can at least check out the Table of Contents. You might find some answers there?

I also discovered an interesting history behind this product. Originally developed by a company called ScanSoft, they merged several years ago with Nuance Communications -- which was focused on providing similar technology to customer call centers. I recently called Sprint and had a machine ask me to "describe" the nature of my call; after I did so, the machine accurately directed me to the right department. To me, that is a far more fascinating use of the technology. Not incidentally, all the reviews I read indicated that DNS 9, the first version released since the merger, far exceeds the capabilities of earlier versions.

Neat stuff.

[bwelford]

Ron, you put me to shame.

In the press of events, I run it almost 'out of the box'. I should have done the research you're pointing to and will do so when I can. The actual version I'm using is Dragon Naturally Speaking version 9.10.000.141 SP1 B, i.e. the Standard version. Even so it works surprisingly well.

[A.N.Onym]

Barry, how much time does it take for you to write a good, lengthy post (ala for your newsletter) with DNS, now that it is trained properly?

Writing has always been the bottleneck for me, because it takes so much time on large scale projects. Maybe speech to text software is a solution to my problem.

(So no, I don't talk to my blog, but I'd like to, if it helps my work.)

Edited by A.N.Onym, 06 May 2008 - 09:22 PM.

[bwelford]

You raise a very interesting point, Yura.

If you're comfortable in giving an off-the-cuff speech, then this voice technology really works well for you. With only an outline of the way the logic flow may go, you just talk as if you were talking to someone else. It's best to be a little more deliberate, but you can get an excellent first draft this way.

It may even be better than if you type. I do type fairly fast as it happens, but nevertheless the mechanical action of your fingers does occupy a small part of your attention, however good you are. Just talking is very much easier.

[Respree]

I think what's even more impressive than talking to your blog is having your blog talk back to you.

I remember seeing a program called A.L.I.C.E. a couple of years ago. Just revisited the site a couple of minutes ago and boy its that entertaining.

I think they call them 'chatbots'.

Have a look (click on the leftmost link that says "Chat with Alice").
http://alicebot.blogspot.com/

Would love to hear from anyone who knows how this works. Fascinating.

[A.N.Onym]

I type pretty fast, too, but compared to speech, still a bit slower. And much slower, than I think, too =)

And yes, talking is more natural, than typing. Though editing is what scares me. Might make a good speaking practice, too.

Still, have you compared how much faster it takes you to talk a post up instead of writing? Basically, that's the only reason I can consider DNS for now.

[bwelford]

I haven't done any precise time measurements but I'm sure if I did the saving would be considerable. I would guess a 50% reduction.

The other advantage is that you can relax. It's no longer like working.

<later edit>
@Respree That A.L.I.C.E. site could be addictive. I'd never seen that before. Off course this is all Off-topic. LOL

Edited by bwelford, 06 May 2008 - 10:41 PM.

[A.N.Onym]

Thanks, Barry. I guess I now need a good microphone. Maybe something the company will invest in =)

Edited by A.N.Onym, 06 May 2008 - 11:34 PM.

[kulpreet_singh]

This is so cool I didn't know it was so easily available and that the technology was actually good.

[bwelford]

I must admit Ron's earlier comments about ways of getting more information about Dragon Naturally Speaking pricked my conscience a little. In a busy life, sometimes if it works out of the box, you just use it. After all, real men don't read instructions. .. and it really does work pretty well.

So I've done a little more research, which may be helpful to others contemplating using DNS. In fact, the reviews that Ron mentioned are more often than not positive. However, I've wandered around the help screens and options provided by DNS and have improved the performance mightily, even above what I was seeing.

DNS will take dictation and will also control your computer. The default is that both are always on. I have now switched to using only dictation mode. This seemed to produce a sizable improvement in speed. Another thing that helps my efficiency, I believe is using Metapad, rather than the Dictation Box provided by DNS. Metapad is a beefed up version of Notepad with some improved methods of formatting text. It also can use live hyperlinks. I find this combination works very well for me. Once I have prepared my text in Metapad, I then cut and paste it into the application, for example, this particular Forum post. I've also tweaked some of the other options and can now largely dictate without needing to correct the text.

So count me one happy camper. Here is a product that scores 9.5 out of 10. Having said that, perhaps I'm now allowed a rant. Ron pointed out that the DNS service is pitiable. Unfortunately that is so true. Like so many other expert high-tech companies, Nuance, which acquired ScanSoft, the creators of DNS, is entirely product driven and shockingly non-customer-centric. Think Microsoft or Google. My most extreme example is Intuit Canada, however that's a story for another day.

In my zeal for improvement, I decided to use one of the Tools mentioned in the Accuracy Center. This is the Acoustic and Language Model Optimizer. Since it mentions that it takes over the whole CPU, I left it running overnight. This morning I found the following Error Message: 0x80040154 class not registered. I went to the somewhat quiet DNS Forum that Ron had mentioned. The problem was mentioned several times, and the only solution mentioned in the forum seem to be linked to some Microsoft plugin that I was missing. However KnowBrainer, one of their VARs seemed to have his own Forum, and there I found the KnowBrainer Quick Tips (PDF file). This gave me the following Tip.

Query
Running the Acoustic Optimizer Produces Error Message: 0x80040154 class not registered
Answer
The simplest solution is to not run the Acoustic Optimizer because it's no longer needed. DNS 9 introduces a new "SilentAdapt" algorithm which updates your Acoustic Model as you dictate. As long as you save your corrections, your acoustic model will be updated automatically and it's no longer necessary to run the Acoustic Optimizer.

Thanks, DNS, why are you still suggesting the use of this tool, when it is no longer needed.

My still remaining puzzlement is on the whole training aspect of such voice technology. As I mentioned, it still occasionally guesses wildly wrong, but puts the correct alternative at #2 or #3. In fact, I had only a handful of errors in dictating this post, and almost all of them were like that.

Perhaps, Garrick, I need to do some more exploration of A.L.I.C.E. I'm somewhat intrigued by another similar site called Pandorabots!. Has anyone else experience in this area?

[kulpreet_singh]

So - as the typical "what's the bottom line" impatient web reader, I have to ask... if I would like to "talk to my blog," which application would you guys recommend?

Edited by kulpreet_singh, 17 May 2008 - 07:27 PM.

[bwelford]

Dragon Naturally Speaking is doing a good job for me. However it is important to be sure your microphone is adequate and then do some work to optimize the settings (via the Accuracy Center, which is part of DNS).