Dragon NaturallySpeaking 3.0 Professional versus IBM ViaVoice98 Executive Edition
Natural speech recognition is certainly one of the most alluring promises of modern computing. Perhaps we've all watched a few too many Star Trek episodes to have a realistic expectation, but the latest generation of speech recognition programs is finally beginning to achieve the level of usefulness and convenience we think most people will enjoy and use regularly. In fact, we dictated this report using the programs.The leaders in this new generation of speech recognition software are IBM, Dragon Systems, and the Belgian company Lernout & Hauspie. IBM, which has spent over 25 years researching natural language recognition currently offers a family of products known as ViaVoice. Dragon Systems has also been in the speech recognition business for number of years and, like IBM, originally offered programs that required the user to speak -- in -- discrete -- words -- like -- this. However, the newest offerings from these companies now offer what is known as continuous recognition. In other words, you can simply say the words you want to dictate into a microphone (typically supplied with the product) and the software interprets the words and translates into text.
As you can probably imagine, this is a massively difficult task for a computer and, as such, requires a fairly powerful PC to achieve successful results and adequate performance. We tested the IBM and Dragon packages on a 180 MHz Pentium PC and a Pentium II 233 with good results. However, we could not get the software to work properly on a Pentium 90; a Pentium 133 was marginal.
We also found that the sound card in a system was critical to the quality of speech recognition. On the portable computer systems we tested the IBM and Dragon software on, the setup programs warned us that the sound card we were using was not adequate for acceptable performance. In these cases, and optional battery-powered adapter should be used to improve compatibility. Dragon recommends a device known as the VXI PARROTT Translator for universal compatibility with all PC sound cards. See www.vxicorp.com for details.
So, How Accurate Is It?
So, how accurate is speech recognition? Well, everything you have read so far has been dictated and corrected and punctuated entirely by Dragon NaturallySpeaking. Moreover, the software can be used to format text by simply saying "italicize that" or inserting similar commands for bold face, centering, bullet lists or even typeface selections.
Some people find fault with Dragon NaturallySpeaking for the length of time it takes to start. Indeed, on our 266 MHz test machine, the program took almost 30 seconds to start up and initialize its vocabulary and input window. However, once running, NaturallySpeaking 3.0 adds speech input capabilities to virtually all Windows programs. We were able to dictate responses to email in the look Express and speak directly – albeit somewhat sluggishly—into Microsoft Word. (The Professional version of the program also includes a speech-enabled version of Corel WordPerfect.) It’s remarkably easy to fix mis-recognized words in NaturallySpeaking. You simply say "correct that" and a list of alternate spellings pops up. If the correct spelling is the second item in the list, you say "choose two" and it’s fixed. You can train a new word here, too, and the program will learn it so that it doesn’t make the same mistake next time.
IBM ViaVoice
Both IBM ViaVoice and Dragon NaturallySpeaking include the ability to issue voice commands-a word or phrase that causes the program to take some action other than typing the text. voice commands may thus be thought of as macros.
ViaVoice also has the ability to speak your text with anyone of several synthesized voices, optional accompanied by an animated on-screen character. like NaturallySpeaking Professional, ViaVoice executive edition can dictate directly into virtually any application.
ViaVoice took about half the time to start up the NaturallySpeaking required. We also liked the convenient on-screen toolbar that ViaVoice uses to provide access to tools and commands.
However, despite the many improvements IBM has made in ViaVoice 98, we found ViaVoice to be less accurate than NaturallySpeaking, often making egregious errors such as inserting the word "police" when we said the word "voice." ViaVoice also tended to insert spurious words, such as "a" and "of," fairly frequently.
ViaVoice was also particularly poor at spelling out words using the "spell out" command. For some reason it insisted on putting spaces between the characters of any words we attempted to spell out -- for example, V X I P A R R O T T. If there's a way to disable this, it wasn't obvious from the documentation.
Its grammatical skills, too, were lacking. For example, ViaVoice inserted an apostrophe in the word "its" in the previous sentence and it didn't have enough sense to spell the word "too" correctly. In short, we found ViaVoice too frustrating to use instead of typing. NaturallySpeaking, on the other hand, managed these sorts of problems more intelligently. It's not perfect either, but NaturallySpeaking didn't foul up in either of these cases.
With NaturallySpeaking Professional, you can just say "new Paragraph" to skip a line
It's not foolproof, but works pretty well. You can also say "press tab" to invoke a tab function -- useful in database applications. To enter an unknown name, there are several ways to enter the characters. You can say "spell that: A B C" or you can use alpha bravo charlie. To spell a word (say, the name "Dmytriw"), you can say "Spell that: type shift-key D m y t r i w... press OK" to spell the word. The system already knows most common names and you can train it so that it will not get it wrong again. For example, it used to think my name was spelled "Graham."
It's our recommended voice recognition solution at this time.
Post new comment