OS/2 e-Zine! - VoiceLearn v1.0

VoiceLearn v1.0 - by Martin R. Hadam

The Vocabulary Trainer for OS/2 VoiceType

Have you ever considered giving up on VoiceType just because every other word needed correction? Or because you were tired of stop-and-go? It didn't perform like this because your pronunciation was bad or you did something wrong; your vocabulary just didn't match the one provided with the installation. When you checked with IBM they would not sell you a professional, pre-made vocabulary simply because those are only available for extremely few topics (like radiology, emergency medicine, journalism, etc.) and even fewer if your native language is not German or English.

What did you do? Keep up the good spirit and go all the way through all the pain? Some did and ended up with a splendid personal VoiceType vocabulary, perfectly suited to their needs. On the other hand, my guess is that many of you only kept VoiceType in your tag-line or used it only for dictating e-mail. Hardly any people use it professionally without add-on vocabularies.

Rejoice! As of today, relief is available: just order your copy of VoiceLearn (GIF 17.1k) (available in both German and English) from ROM LogicWare, makers of the venerable PAPYRUS word processor for OS/2, which beats the competition by being the first and only word processor to accommodate direct dictation and formatting using OS/2 VoiceType.

What I Like About VoiceLearn

What will VoiceLearn do for you? In brief, VoiceLearn will read text files and extract all words unknown to VoiceType. It will present them to you as an editable listing and request pronunciations for each selected word. Finally, VoiceLearn will store those words with your personal vocabulary and update the associated trigram statistics, resulting in a much enlarged and improved personal vocabulary.

What does this mean to you? Assume you're a student and supposed to write a paper on topic "X". The topic is new to you and VoiceType, but some colleagues of yours have written papers on this and related topics in the past. Using those papers with VoiceLearn will enable you to use VoiceType with optimal recognition rates on every detail of topic "X" right from the start.

Do you intend on dictating your thesis? Are you a scientist? You can grab the vocabulary of an entire scientific discipline from the appropriate CDROM of a standard textbook or multi-megabytes of text perfectly tailored to your needs from Medline or other services' abstracts. (Please note however, that many sources of text data are subject to Copyright regulations and act accordingly.)

If you are a business professional, grab all your previous correspondence and feed it into VoiceLearn: almost immediately after, VoiceType will be knowledgeable about your business and know all the words you are using.

As a medical specialist, do the same with your patients' records and start using VoiceType, now, all of a sudden, highly specialized into neurology's, pediatrics' or ophthalmology's (you name it) terminology, etc.

If you're into translation, you may notice that VoiceLearn supports all language versions of VoiceType out of the box. Thus, using a single installation of VoiceLearn, you can update all your personal vocabularies in a multi-language installation of VoiceType using the procedures mentioned above.

Installation and Initialization

Your personalized copy of VoiceLearn installs easily into any directory. Since it will automatically hook into the current default VoiceType registration, you may have to choose the appropriate one before using VoiceLearn, though.

On opening, you are prompted with a notebook in classical style, which is nice since you can read through the tabs without scrolling. First you select (GIF 11.2k) your text files (wildcards are allowed), which should be in straight ASCII. Then you decide on any characters (GIF 14k) which may or may not be part of a "proper" word. (This is required since VoiceLearn takes everything separated by spaces as a separate word and punctuation has to be taken into account.) While this may appear difficult to you, the default settings usually cover almost everything, so there is no need to worry. You may then decide what to do with text enclosed within various kinds of parentheses (GIF 10.5k). (This and the preceding settings can be used to read HTML or SGML based text straight away.) If you're using formatted documents (GIF 13k), you can define which parts should be skipped during evaluation. And finally, you may select the proper codepage (GIF 17.4k) for your documents to correctly display the special characters in your language (like umlauts, accents etc.). Off you go!

Learning

When the progress bar has reached its end, which takes about 8 minutes for about 7MB of text on my 133MHz Pentium, you are prompted first with statistics (GIF 8.7k), and then with a list box containing all novel words listed either in descending frequency (GIF 13.9k) or alphabetical (GIF 14.3k) order. It is advisable to define a cutoff for the minimum frequency (a good value is 5) in order to (i) keep the list at reasonable length and (ii) maintain good trigram statistics for each newly added word. Please note, however, that you may come up with more than 50,000 new words in total without much effort!

Once finished editing your list, you'll have to press "start" and concentrate on the neat traffic lights used for dictation: if green, you just pronounce the word displayed in the entry field below. VoiceLearn will automatically determine the end of your word, and go back via yellow to red for processing voice information. Yellow again will alert you for a new cycle and there it goes...

Dictation this way is fast and easy; if you wish, you may define your own pace by pressing the spacebar though. Finally, before exiting the program, VoiceLearn will quietly update the trigram statistics of all newly added words. On the next start of VoiceType, it will operate just as if you'd never dictated anything other than the words you just added. It's amazing to me every time I see that!

Other Notes

From my own experience I'd recommend to collect as much text as possible and process it all in one run or a few major ones. Even though this may result in a significant task at one time, you are rewarded with an optimal vocabulary based on the increased word frequency and resulting improved trigram statistics.

This has nothing to do with VoiceLearn but rather with the way VoiceType operates. Of course, if only small amounts of text are available to you, you'll have to live with it. It will still be worthwhile to run VoiceLearn. In such cases, however, you may also decide to look at words of lower frequency and consider that the trigram statistics will only be based on a few occurrences.

Backing up your basic VoiceType enrollment is also highly recommended and you may decide to keep different specialized vocabularies for various tasks.

Complaints

Any complaints? Only gripe is that I've been waiting for something like this for years... Seriously, the program is that good!

Conclusions

In summary, VoiceLearn is a must for the professional VoiceType user in any of the supported languages. It works with VoiceType 3.0 for OS/2 and also with the special Aptiva VoiceType Preload v1.2; it will not operate on other VoiceType 1.x versions. VoiceLearn is easy to use and replaces the need for costly pre-made vocabularies. It works flawlessly and is worth every penny it costs. Highly recommended.

VoiceLearn v1.0
by ROM LogicWare
MSRP: US$298

Dr. Martin R. Hadam is in the Department of Pediatrics at the Medical School in Hannover, Germany. His research focusses on inborn errors of the immune system. He has used OS/2 since v2.0 and has been using IPDS (later dubbed VoiceType) from its very beginning in both the US and German versions. He also owns one of the few German VoiceType v1.2 preloaded Aptivas in existence (which, sadly enough, never made it to the market).

[Our Sponsor: Perez Computing Services - Makers of Ctrl-Alt-Del Commander and IPF Editor.]

This page is maintained by Falcon Networking. We welcome your suggestions.