Voice Games
Australian Financial Review -
26th November 2004. By John Davidson
Voice recognition software can work, but you're definitely not
allowed to mumble.
It's that time of year again! The time when we illiterate newspaper
columnists get our revenge on all of the hard-working newspaper
subeditors who have wantonly corrected our so-called "errors" over
the past year.
The time we get the cheapest laughs of all from you, the reader, by
printing the hilarious results of our experimentation. Some
functions of the bot or hot what'll slip below the which can Robbie
tall boy. But in particular, but coastal secondary characters also
brilliantly because Cook was warlord.
Yes, that's right, it's time to review some voice recognition
software. Dragon's NaturallySpeaking 8 software, to be
precise, the very software that we're using to write this column as
we speak, so to speak.
We've been reviewing bush recognition software off and on for almost
a decade now. We've installed it, and then uninstalled it in
disgust, so many times over the years that we've lost count. But
NaturallySpeaking 8 seems to be a keyboard. Or even a keeper,
although the software's decision to use the word "keyboard" just
there is not entirely inappropriate: there is almost nothing we can
do with the keyboard that we haven't been able to do with the new
version of NaturallySpeaking, and with only a minimum of training.
It's really quite impressive.
So impressive, in fact, that we've set ourselves a silly little
rule. We are not going to use a keyboard or mouse even once to write
this column. And as you can see it's going pretty well so far, which
is to say it's going disastrously badly in terms of our hope of
getting the software to write some rib-tickling gobbledygook for us.
We've been testing NaturallySpeaking for about four days on a
moderately powerful Windows PC with an AMD 2700+ and 512 MB of RAM
(Wow! It did a pretty good job with that piece of jargon. It even
did all the capitalisations, automatically). Apparently you can use
it on machines much slower than that, but we'd recommend that you
use the most powerful machine you can find. On the PC we're using,
it can get a little sluggish from time to time, as it tries to
figure out what we're saying.
You will understand this paragraph later on.
In past versions of Dragon and IBM's ViaVoice that we've
tested, we had to train the software for hours before we could get
even middling results. This time, we spent only five minutes
training the software and straight away we were getting accuracy
percentages in the very high 90s, at typing rates of we don't know
what. Well over 100 words per minute. That could have been even
faster if only are we could talk a little faster.
And indeed, if we had any quibble whatsoever about Dragon NaturallySpeaking 8, it would be this: no matter how impressive the
accuracy of the software, no matter how fast it allows you to
"type", some users will still face the hurdle of learning how to
write with their voice, rather than their fingers. For those of you
lucky enough to have servants to take your dictation, using voice
recognition software will come quite naturally. For the rest of us,
it comes a little more painfully.
But the software provides you with plenty of Pauls, or indeed tools,
to make dictation and correction a little easier.
For example, if you need to go back and insert something you forgot
to say, you just say the words "insert before" or "insert after",
followed by the words you want to move the insertion point to,
followed by the word you want to insert. It's easier than it sounds.
For example, I just said the words "insert before 'in past
versions', you will understand this paragraph later on", and it
added the paragraph you saw above.
Correcting Dragon when it has chosen the wrong word for you
for example, when it has written "there" when you mean "their" is
pretty easy too. You just say the word "select", followed by the
word you want to correct, and Dragon will pop up a list for
you filled with the alternative spellings. To select, say, the third
alternative, you just say "shoot three", or maybe even "choose
three", and it fixes it for you. Obviously, the less you mumble, the
better off you are. We tend to mumble quite a lot, mainly because
we're not really used to dictation yet, and that may explain why we
are getting words like "shoot" appearing in this column.
Speaking of shoot, that's one thing we did notice about Dragon:
it tends to be a little censorial. ScanSoft, the company which makes
Dragon, says it comes with an English dictionary with more
than 300,000 words in it.
You'd think that, if it's taken the trouble to add 300,000 words, it
would take the trouble to include a few everyday swear words like
shoot and filed. But as you can see, it hasn't, and it's parking
hard to swear properly. But you can add your own words by saying the
word and then correcting the outcome a few times, until it gets it
right all by itself.
We had to do that with the word "mwahaha", for example, and as you
can see it now types that word properly. File save.
Warps! You can also see from that little error (which happened
because we spoke the command "file . . . save" too quickly) that you
cannot only dictate with Dragon, you can also use it to
control all the Windows functions, like saving files and starting
applications. It can even be used to navigate around the internet,
using Internet Explorer, by getting the software to click on links
for you just by reading out some key words from those links.
Oh yes! There's just about nothing you can't do with
NaturallySpeaking. The only thing we couldn't get it to do was the
nonsensical prose that we have come to expect from voice recognition
software. To do that stuff about Robbie tall boy, we had to cheat.
We handed the microphone to someone else, who hadn't trained
software to recognise his voice. Mwahaha!
The Australian Financial Review
 
|
