|
tranfree issue 25 - 5th February 2001
What's All The Talk About?
Where To Spend Your Money In Speech Recognition
by Michael Benis
Why Bother?
The why of speech recognition is very simple...
most of us can talk faster than we can type
...so it increases our productivity. In addition, the fact that
you can move around because you're not tied to your keyboard
helps keep you feeling fresher even when you're under pressure.
As a result, you can achieve higher productivity for more hours
in the day without falling to pieces. Perhaps most important of
all...
it's great protection against (RSI) repetitive strain injury
...and one of the few sure-fire ways of getting over it.
Does It Work?
Yes, but you have to get everything right...
- your computer
- sound card
- microphone
- speech recognition product
- how you use the system
Go through every step methodically and you'll end up...
 ...grinning
from ear to ear as you lean back in your chair, feet up
on your desk merrily chatting away to your computer.
But don't rush things. It will take you around a month before
you're using your system with complete success and confidence.
Although you'll find more than enough useful pointers in my old
ITI Bulletin articles on the translatortips.com site, there
really is no substitute for the certainty, speed and peace of
mind you'll enjoy if you can actually find someone to give you
first-hand advice and one-on-one training. But be careful...
All Experts Aren't Alike
Translators aren't typical of most speech recognition users. Not
only do we generate large amounts of text, but the vocabulary and
style of the texts we dictate vary greatly.
As a result, choices that may be acceptable to less demanding
users are very much less acceptable where we are concerned. So
ignore what you read in the high street magazines. If you opt for
training, go for someone who has been recommended by a fellow
translator. Above all, follow this advice...
The Basics
To be really happy, you want at least...
- 500 MHz processor,
- 128 MB RAM
- Turtle Beach Montego or Sound Blaster Live! soundcard
If you're buying new, get as close as you can to doubling those
specifications (processor speed and RAM) before the price jump
per MHz starts getting silly.
Dragon NaturallySpeaking (Version 5)
NaturallySpeaking was the first continuous speech recognition
system on the market and still leads the field. It's the easiest
to use, gives you the highest recognition accuracy and is the
only system that allows you to correct entire sentences at a
time.
What's more, it offers a whole range of commands that make
dictation, correction and navigation faster than the competition.
Unfortunately, the latest version is not an improvement in every
respect, although Dragon is working to sort things out.
IBM ViaVoice (Release 8)
ViaVoice is close on the heels of NaturallySpeaking for
recognition accuracy. But for ease of use, stability and
productivity when dictating long texts, it definitely trails in
second place.
That said, it remains NaturallySpeaking's only serious competitor
and unquestionably offers better value for money in its
Professional version.
Lernout & Hauspie Voice Xpress Pro (Version 5)
Although a considerable improvement on its predecessor, Voice
Xpress quite simply isn't a serious contender for translators.
The main reason is that it won't play back what you said. (All
the other products do this like a dictaphone).
Since speech recognition errors can be very different from what
you actually said, you'll spend a lot of time scratching your
head trying to remember with this product. That's wasted time in
which you could have simply got on with the job.
Philips FreeSpeech 2000
An attractively-priced product, offering a choice of languages,
the ease of use and recognition accuracy of FreeSpeech 2000
simply don't compare with NaturallySpeaking and ViaVoice.
It also suffers the disadvantage that you can only play back and
correct what you said in a series of separate operations,
although you can do this very quickly if you invest in the
splendid SpeechMike Pro.
Where To Put Your Money
If you're already using IBM ViaVoice and are happy with it, then
you'll be delighted with the upgrade to Release 8.
If you're using Dragon NaturallySpeaking Version 4, hang onto it
and wait for Dragon to iron out the bugs in Version 5.
Likewise, if you're looking to buy your first system, search out
one of the remaining copies of Version 4. You'll save yourself a
tidy sum and will be able to upgrade to Version 5 when it
actually represents an improvement in every respect. But don't
worry if your dealer can only lay their hands on Version 5, it's
still the best choice for translators.
Don't forget that you'll need the much more expensive
"Professional" version of NaturallySpeaking if you want to
integrate speech recognition with Translation Memory.
That's particularly true of Trados, for which you have to write
macros that drive the menus if you want to avoid damaging its
tags. With any of the other Translation Memory programs on the
market, you can obtain the results you want by just dictating
their keystroke shortcuts in NaturallySpeaking Preferred.
Good Luck And Let Us Know How You Get On
So you've decided to treat your hands to a lazy 2001. What next?
- Go through the enrolment procedure slowly and methodically, but don't do much more than the bare minimum
- Get used to dictating for several days, noting which style works best
- Then enrol again
This time, read several texts and feed as many old translations
as possible into the "Vocabulary Builder" or its equivalent.
After a month or so, let us know how you got on.
mb@tranfree.com
The more feedback we receive the better the information we can provide.
Michael Benis - Freelance Copywriter, Interpreter, Journalist
Translator of French / Italian / US -> UK English
EC Freelance Translator - Approved Police Interpreter
Click here to return to
tranfree 25 main page
| 