Talking yourself out of trouble

 

Update on speech recognition: Dragon NaturallySpeaking for French German Italian and Spanish; IBM ViaVoice Millennium Pro, Philips FreeSpeech 2000 and Talkback 2000.

 

Funny title, you'll say. Not to me, it isn't. I spent this week not so merrily working my way into an early grave. I'm exhausted. But I've met all my deadlines and I'm not a broken man. That's a minor miracle and there's only one reason for it: speech recognition.

 

Now, don't worry, this isn't a preamble to repeating everything I told you earlier on this year. Suffice it to say, if there's one tool I use to make my life as a translator and writer more productive, more comfortable and generally more enjoyable that I would quite happily commend to you all, this is it. What's more, not one of the many ITI members who contacted me after the comparative review came out earlier this year and started using speech recognition as a result has regretted the move. Once you master it with a decent product, you'll increase your productivity and significantly reduce the strain to which you subject your body. Your fingers, hands and shoulders will be relieved of keyboard strain. Your back will benefit from being able to move about as you work. Your eyes will be happier because they spend less time staring at the screen. And you, of course, will be a lot happier as a result of it all.

 

Let me give you some concrete examples.

 

Antonio Aparicio, our Bulletin editor, was so impressed with the Spanish version of NaturallySpeaking that he decided to use it with Déjà Vu to work on the new version of a manual he'd translated earlier in the year. Within a matter of days (for which read two) he was achieving productivity of a thousand words per hour. That's way over anything he's likely to average for the year, but you've got to admit it was a stunning start!

 

Similarly, Ros Schwarz started using NaturallySpeaking just a couple of months ago. After her initial learning curve, she is now finding that what used to be a day's work translating novels now gets done in many hours less, and with less physical stress, too.

 

Don't forget that learning curve, though. All of these systems need to learn how you speak and whether you like it or not, you'll make subtle changes to the way you speak that gets you better results. More than anything else, you'll need to change the way you work if you've never dictated before. Persevere, however, and you'll never look back. And - to be alarmist for a moment - you can also relax in the knowledge that if you're laughing all the way to the bank about your income now, you'll be able to carry on smiling in the future, since the chances of developing repetitive strain injury will be very much if not totally reduced. High earners who bash out all their work on a keyboard just can’t be sure.

 

Well, now that I've done my evangelical bit we can all get down to the specifics of this review. Our aim here is partly to keep you up-to-date on the best products available out there, but also to answer the big unanswered question that arose out of the comparative review earlier this year.

 

 

What about my language?

 

Dragon NaturallySpeaking 5.2 - UK English, French, German, Italian and Spanish

 

Yes, that was the big question, of course. This stuff may work well in English, but can it handle French, German, Italian and Spanish? Well, to be quite frank the list of languages was longer, but that's what Dragon very kindly allowed us to test so that we can now give you an answer. What's more, they pulled out all the stops to bring us all these languages in the new version 3.52 (currently about to be superseded by version 4.0), meaning I could also talk about what sort of an improvement this made on version 3.0 that had been reported on in the comparative review.

 

Having a perverse nature I'll take you through all that in reverse order. The arrival of Dragon NaturallySpeaking 3.52 in my computer confirmed the program's position as king of the castle (tested in PCs with a 233 and 500 MHz processor). Training is easier than its competitors, initial dictation accuracy higher, learning faster, correction quicker and navigation easier, too. The new version also featured a number of enhancements that Dragon out of some sense of strange natural modestly had kept quiet about, not least of all the fact that it now works slightly faster overall, but particularly in Word and for navigation. Recognition accuracy was also improved but most important of all was the fact that one could now easily navigate around windows by voice.

 

As a result of this improved navigation, all of you who are not interested in the wonders of Translation Memory can quite happily buy Dragon NaturallySpeaking Preferred and forgo the macro capabilities offered by the much more expensive Professional version. This difference in price is the biggest fly in Dragon's ointment. I don't doubt they've lost sales of both versions as a result of it.

 

As an example of what you can get away with, Antonio's achievement mentioned above was using Preferred. He's not at all convinced that there's any point in him spending double the money. The only way he'll be convinced is to try it and see. Because it really does make a massive difference being able to fully integrate the two, and you will, in fact, see a return on your investment within a matter of weeks. That's good news, of course, but doesn't really make NaturallySpeaking Professional any better value.

 

Whether you get Preferred or Professional, however, here’s a quick list recapping the features that make productivity so much higher in Dragon: Firstly, initial recognition accuracy is higher than its competitors. Secondly it's less affected by noise and breathing. Thirdly it seems to learn quicker (I was recently able to confirm all three again recently when software problems while testing another product meant that I had to re-train NaturallySpeaking from scratch). Fourth, it has superior document navigation commands, allowing you to insert the cursor at any point by simply saying "insert" followed by the preceding or following word, and to select a chunk of text by saying select "first word" to "last word". You can also delete or correct a chunk of text using a single command in the same way. This makes editing and correction significantly faster. Another very nice feature is that if you disagree with a translation halfway through a sentence you can simply say "Resume with" followed by the last word you want to retain to be able to immediately continue dictating without having to select or delete the recently dictated passage you've decided you want to replace. Fifth, correction is faster because it's a one command process, because you can edit in the correction window using all your usual voice commands, because spelling in the correction window works better than any of Dragon's competitors and because the list of alternatives from which you can choose is more likely to contain what you actually said than any of its rivals. Making this even simpler, there's an option that makes Dragon play back exactly what you said when you make a correction so that you don't lose any time trying to remember what it was.

 

In short, Dragon's NaturallySpeaking is still the solution that will net you the biggest productivity gains and least frustration. I heartily recommend it. If you've already replaced your computer, serviced the printer, cleaned the fax, put duct tape all over your dictionaries and resolved to spend less on mince pies this year, then drop Santa a note about this. Just make sure he brings you version 4.00, however. We haven't tested this yet, but it features twice the active vocabulary as version 3.52 and a number of other very significant enhancements, including user training that has been reduced to five minutes. What this big active vocabulary means is that there are twice as many words all waiting there waving their hands about just dying to hear you say their name. As a result, whatever you say is even more likely to be recognised and will be recognised even quicker. Well, that's the theory anyway and the reports from the USA and Germany (where version 4 is already on the market) are that it very definitely works in practice. Dragon have assured me that a copy is winging its way over to the ITI Bulletin as I write this, and we'll bring you a full report as soon as it does.

 

 

But does it work in my language?!

 

Congratulations for having skipped straight here if this was the question to which you most wanted an answer. Here it is: Yes.

 

I myself tested the Italian version and if anything the results are actually better than for the English, which isn't surprising since Italian has the very good grace to be written pretty much as it sounds.

 

There isn't really any more for me to say about it than that. Everything I've said about the English version applies to the Italian version, too. At the moment Dragon NaturallySpeaking is still the standard by which I would measure all other speech recognition programs. The Italian version offers no less than the English one and recognition, as I say is higher. So, with nothing to add on that score, all that it remains for me to go on about is what you need to bear in mind if you want to use more than one language version on the same machine.

 

Installation isn't a problem because Dragon recognises you have another language version present and advises you to use another installation directory. Follow that advice! The default directory that Dragon chooses is NatSpeak (which is where my English version was installed), so I simply added an It to get C:\NatSpeakIt for Italian. You can, however, call the directory anything you like, such as "I like strawberry yogurt", for example. What you mustn't under any circumstances do, however, is install NaturallySpeaking in your root directory (C:\ as opposed to, for example, C:\I like strawberry yogurt\). Dragon will work perfectly well if you do this, but should you ever decide to uninstall the program and tell it you don't want to save your speech files, it will merrily delete a number of crucial program files as well, killing your Windows installation.

 

So, if you already have another language version of NaturallySpeaking on your system, you install your second (or third!) NaturallySpeaking in a directory with a different name to either C:\ or C:\NatSpeak\. What you don't do during the installation is to check the little option that gets Dragon to install NaturalWord - unless you want to use the new language for the NaturalWord commands in Microsoft's Word. The problem is that NaturalWord support can't actually handle more than one language. This mean you can dictate into Word with any of the language versions you have installed using NaturalText, but only get all the various NaturalWord formatting commands for one of them. This may be a problem for those of you who regularly work into more than one language. For me, the choice was easy since I only work into English and use Italian for correspondence. When I first installed NaturallySpeaking in Italian I installed NaturalWord and ended up with the Italian version, which looked for and used the Italian program. Having realised what had happened I simply reinstalled the English version of NaturalWord, which overwrote the Italian (and activates the English version of the program). One final tip: NaturalWord for let us say English won't work if you were using NaturallySpeaking Italian immediately before it. You'll need to open NaturallySpeaking English first.

 

Something else you need to bear in mind is how the system will perform when your operating system is for one language and NaturallySpeaking for another. I, for example, found that using NaturallySpeaking Italian in Windows 98 English, I could quite happily navigate around the Start button menus but had to dictate keystroke shortcuts to activate the menus in English language programs (such as Word for example). The Italian commands for resizing and closing Windows all worked quite happily, however. It seemed simple. I couldn’t say English commands when in Word or any other (English) program. Or at least that’s what I thought until one afternoon in a fit of perverse humour I tried speaking the English commands for Word’s menus in an uncompromisingly Italian accent. A complete success! Everything worked perfectly!

 

I only have one real gripe and that's about spelling in NaturallySpeaking Italian. One of the very convenient things about NaturallySpeaking is that you can spell things out letter by letter and it actually works. This makes correction much quicker. This all works fine in NaturallySpeaking Italian except that no one thought of making it possible to dictate accented characters as well. This was a double nuisance for me because I don't have an Italian keyboard any more. Dragon, please take note!

 

 

French frenzy

 

French was the crunch language as far as I was concerned for the simple reason that the singular and plural forms of words often sound the same. Making things even more challenging, this difficulty is compounded by little Romance language gems such as the preceding direct object rule. Unfortunately, a number of problems meant that our French tester had been using the program for less than a week as we went to press, so there isn't really enough to report. She certainly thought it would be workable and was impressed at the speed with which it was learning, but sadly we are still at too early a stage to confidently state anything one way or another. We'll keep you updated on this.

 

 

German on the edge

 

The main problem experienced by our German tester, Anke Vogelhuber, was insufficient RAM. Anke's machine only had 64 MB which is the minimum recommended by Dragon and insufficient to make full use of Dragon's BestMatch technology and NaturalWord. It also meant the system was quite slow. She had no previous experience in dictating translations, apart from the unsuccessful trial of a competitor product (ViaVoice 98) which she found had insufficient recognition accuracy even after full enrolment and was too sensitive to breath noises to be really usable.

 

As with the other language versions, the make-or-break period for Anke’s initial learning curve was about a month, during which time the system learned more and more about how she speaks and she learnt how not to slur or clip her words while speaking naturally. Typical problems during this period included missing pronouns like "uns" and "es" or the beginnings of words, such that Anke would get "bei" instead of "wobei". Equally, similar sounding words would be misrecognised until more contextual statistics had been acquired. Confusing "ich" with "die", "hohn" with "hund" and so on were examples of this. Anke's initial report was that she "found the recognition rate for figures VERY bad." The day after I received her report she rang me to say the system had suddenly started getting them all right. In the beginning, performance improvements are often big steps like this rather than a gradual curve. Compound nouns tended to need more initial training, although Anke, who specialises in legal translations, was impressed with the large number of legal compound nouns that were already in the standard vocabulary and recognised immediately.

 

Spanish hits the stars

 

Antonio Aparicio, our Glorious Bulletin Editor, tested the Spanish version himself. He, too, had little experience of dictation and absolutely none with speech recognition. Likewise, he received absolutely no help with either setting up or training the system. After a long journey during which we waited on tenterhooks while the Post Office routed the package from Brighton to Manchester via Abu Dhabi, Antonio finally got up and running. When he rang me two days later, I imagined it was with a problem and got all ready to roll out my helpful expert voice. I should have known better. He was absolutely delighted, already using NaturallySpeaking with Déjà Vu and a firm convert. Life had not been without its problems, since there was a bug in 3.52 which needed a replacement *.dll  file to be downloaded from the Internet (this was also the case for Italian), but there's no point in going into that now with version 4.00 about to come onto the market.

 

 

A success in any language?

 

That certainly seems to be the case, although we can't yet put our hands on our hearts where French is concerned. The big thing to remember is you need to persevere for the first month or so. When I rang Ros Schwartz to ask if she minded my mentioning her by name, I read her what I'd written above and her comment was: "You can sound much more enthusiastic than that! I'm absolutely over the moon! I was really quite sceptical in the beginning and had to push myself to persevere, but I'm really glad I did now." Ros was using a system set up and configured by GMS, who have now acquired considerable experience in meeting translators’ speech recognition requirements using NaturallySpeaking in particular. Someone offering similar expertise for ViaVoice would be our own ex-chairman Roger Fletcher.

 

Another satisfied GMS customer is Emma Gledhill, who has integrated Dragon NaturallySpeaking Professional with Trados Translator's Workbench and is also delighted with the results. In the beginning she was more concerned about the comfort and quality-of-life benefits she had achieved, since she is a very fast typist and was not anticipating any increase in productivity. Two months down the line, however, all that has changed. Not only is she finding life much less stressful, but she's also getting more work done - or resisting the temptation to work more and enjoying more free time instead. From what we've now seen, it certainly looks as if you can achieve that with any of the languages available for NaturallySpeaking providing you have that initial patience and a powerful enough computer. For full information on all that, refer to part one of this comparative review

 

 

[Panel]

 

Give us a piece of your mind!

 

Please let us know your experiences with any of the speech recognition programs on the market, not only in any of these languages but especially those that we haven't tested, such as Dutch. Considering the significant potential benefits for our health and productivity, that information is too good not to share! You can contact Michael at: michaelbenis@cwcom.net

 

 

Number two comes in from out of the blue!

 

IBM fights back with ViaVoice Millennium Pro

 

For many years speech recognition was a two-horse race: you either went with Dragon or you went with IBM.  Throughout the final years of discrete speech recognition they kept jockeying for position, often leapfrogging one another into pole position (to mix my metaphors). Then Dragon came out with NaturallySpeaking, the first continuous speech recognition product, and in many ways never looked back. In short, they managed to gain the lead and hold it. Well, having languished for a while, Big Blue - as IBM is affectionately known - has now very nearly closed that gap with IBM ViaVoice Millennium Pro.

 

One of the ways they've done this has been to practically double the active vocabulary, the latest trick in speech recognition technology, and one that very definitely works. They've also made enrolment much easier than it used to be, created an "agent" called Woody (who looks like a talking pencil) to help you familiarise yourself with the program, provided full integration with Microsoft Word, Internet Explorer and Windows 98, and increased the mouse control offered by the system.

 

 

Continued here (part two)

First published in ITI Bulletin, 1999.