Voice recognition software

I’m playing with
Dragon NaturallySpeaking again because I want a more efficient means of getting ideas into a more permanent form. When I get excited, I can talk about 5x faster than I type, and I can type 10x faster than I can handwrite. Handwriting doesn’t work very well, especially on a blog (though it does offer some near unbreakable encryption, according to my spouse). DNS claims it can do 3x one’s typing speed.



There are several “levels” of the software. I have the “preferred” edition because when I bought it, I was regularly using Outlook as a mail/calendaring tool. As a special bonus, DNS also works with the “vi” text editor. I don’t have the patience to enter manual editing commands, though.



I’ve run through several of the software’s training passages. Doing so helps it tailor my inflections and improve its accuracy. This is a very tedious process, however David Pogue at the NYT swears it will pay off. (He’s largely the reason I use DNS over IBM’s Via Voice, although both are sold by ScanSoft.).



The software relies heavily on contextual analysis. Thus, as you type, it won’t display until it’s chained together a bunch of phonemes and formed a probable sentence. You have to do this in a relatively quiet space as it does pick up background noise fairly easily. (In my old cubicle environment, this was a big problem.) It’s still not perfect, though, but it shows a lot of promise.

Here are three sample passages and the results of my reading them.




The first passage is some lyrics taken from the middle of the song “You Can Call Me Al,” from Paul Simon’s excellent Graceland CD. First, the original:

A man walks down the street,

he says, why am I short of attention?

Got a short little span of attention,

and whoa, my nights are so long!

Where’s my wife and family?

What if I die here?

Who’ll be my role-model?

Now that my role-model is …. gone …… gone,

he ducked back down the alley,

with some roly-poly, little bat-faced girl.

All along …. along ….

there were incidents and accidents,

there were hints and allegations …..

Here is Dragon Naturally Speaking’s transcription, unmodified except for line breaks for visual comparison:

If a man walks down the street,
he says, why am I short of attention?
Got a short little span of attention,
and whoa, my nights are so long!
Where’s my wife and family?
What if I die here?
Who’ll be my role model
of that my role model is … gone… gone,
he ducked back down the alley
with some roly-poly, little bat faced girl.
All along in its use along in its use
they were incidents and accidents,
there were hints and allegations…

This was better than I expected. It got the dashes between roly-poly, but not bat-faced. It also thought I meant “in its use” instead of “…” (pronounced elipses).



The second passage was taken from yesterday’s blog entry on schedule chicken. I thought this might be a little different because of the long, complex sentences. Here is the original:

If you’ve watched a lot of old movies, you’ve no doubt seen the one where the two hooligans with something to prove decide to race their cars towards each other at high speeds. The first one to turn away, averting certain collision, is the “chicken” and loses face among his cohorts. The other player, referred to as the “winner,” gets bragging rights and access to the fickle heroine.

Here is the DNS transcription. When you read, you have to tell the program where punctuation goes. It’s very unnatural at first. I didn’t know how to tell it to put double quotes around the two words, but otherwise, this is unedited:

And if you’ve watched a lot of old movies, you’ve no doubt seen the one with two hooligans with something to decide race their cars towards each other at high speeds. The first one to turn away, avoiding certain collision, is the chicken and loses face among his cohorts. The other player, referred to as the winner, if bragging rights and access the fickle heroine.

This is not that bad – an added “and,” missing “prove”, “to” and “if” instead of “gets.” I’m still a little new at this, so when I read into the microphone, I’m making more of a conscious effort to enunciate clearly, but I’m still talking at my rapid rate. The program takes upwards of 30 seconds to make sense of what I’m saying before typing the text in the place.



The final passage is excerpted from Fran’s June 18th entry. This is interesting test case because it includes a formal product name and a geographical place:

My proofreading job on Microsoft Encarta ended today. We all went out to lunch and then called it a day. I 

• Share: