most computers that you're likely to buy now can do the job.
IBM's ViaVoice and Dragon Systems' NaturallySpeaking both require a powerful Pentium-based computer with 32 MB of RAM (for Windows 95; NT needs 48 MB), a good-quality sound card, and a significant amount of disk space. I tested them on a 166-MHz non-MMX Pentium with 64 MB of RAM.
Speech software is different, however; it isn't really ready-to-go out of the box. Installation requires time and effort, and you have to commit at least a half hour to read in text and train the software to your voice and pronunciation. I spent an equal amount of time training each product so that my evaluations and comparisons would be fair. And after completing the initial training , I ran the same vocabulary file through each product.
This software presents two distinctly different capabilities: transcription of what you say, and control (using voice commands to
navigate, edit, or format a document). ViaVoice offers only the barest formatting and editing, while NaturallySpeaking has more power -- but only in its own dictation window. Neither package has the flexibility and power of Kurzweil Voice Command for Microsoft Word (see the sidebar "A Commanding Voice for Word").
NaturallySpeaking has been on the market longer than ViaVoice has, and it costs more. So, I had high expectations, and I wasn't disappointed. Every step of the way,
NaturallySpeaking
was better at correctly recognizing what I said. During the testing period I never quite achieved the company's claimed 95 percent accuracy, but recognition clearly improved with time and use.
Performance is highly dependent on the computer's sound card. NaturallySpeaking's initial microphone setup told me that mine was below average and would degrade performance and accuracy. Dragon's Web site lists some tested sound cards, so I installed an inexpensive Hi-Val SounTastic16 ca
rd. The noise-canceling microphone that comes with NaturallySpeaking was similar to that for ViaVoice, but it had a noticeably heavier-duty cord and a black-box battery adapter.
Correction is efficient, and you can select text by either content or form. For example, say "Select [word]," and it highlights that word, or you can say "Select last [or next] word [sentence, paragraph, and so on]. Then you simply say "Correct that" or "Spell that," two modes that are useful in different situations. As with ViaVoice, after correction, the system might ask you to record that word.
With NaturallySpeaking Personal Edition, you can dictate into its own window and then cut and paste text into your word processor. The $695 Deluxe Edition, which was released after this review was finished, integrates with Microsoft Word, supports multiple user profiles and vocabularies, offers text-to-speech and voice recording, and works with DragonDictate, a discrete-speech product, to provide additional command power.
V
iaVoice works fairly similarly to NaturallySpeaking. The key differences are ViaVoice's correction process, its inability to add mass vocabulary, and its greater integration with other software.
When you install
ViaVoice
, it hooks directly into Microsoft Word, giving you a Dictation menu pull-down and toolbar. I also tested a new release of Lotus WordPro that includes ViaVoice. Curiously, this version has an important feature not in the stand-alone product: Vocabulary Expander, which lets you read in text files to increase the system's vocabulary.
Correction is somewhat awkward. The process seems to presume that one per-son will dictate a document and another will edit and correct it. First, you select up to three words of incorrect text using the mouse; you can't use a voice command, as you can with NaturallySpeaking. Next, you click a correction icon or press F2. ViaVoice reads the recorded text and shows it in a correction window.
This can be helpful. For example, my
clearly spoken "feed it additional text" became "Lafayette's additional taxes." You enter your corrections from the keyboard.
Unfortunately, after I selected a word or phrase, the highlighted area sometimes changed, forcing a second correction. For example, I highlighted the phrase "14-inch monitors" and pressed F2 for correction, and the selection area suddenly changed. When I entered the correction, the final text then read "14-inch monitorsrs."
Also, ViaVoice has a feature that NaturallySpeaking doesn't: It can read text in any of several synthetic voices.
Can We Talk?
To provide a simple measure of accuracy and speed, I read a 473-word passage from a book into each test platform -- NaturallySpeaking, and ViaVoice in its SpeechPad, in WordPro, and in Word. I did not look at the on-screen text or correct errors while dictating. I noted the overall time, counted recognition errors, and computed the number of correct words per minute as an overall performance index (
see the figure
for more details).
NaturallySpeaking was faster and made fewer errors. ViaVoice/SpeechPad was a bit slower and less accurate. With either word processor, ViaVoice dictation took twice as long because I was periodically asked to stop dictating so the system could catch up. Each time, I had already read at least 100 words beyond the system's limits. In addition, after I finished dictating into WordPro, for example, the system continued processing for a full minute; NaturallySpeaking needed but 10 seconds.
Both products are useful, but I'd pick NaturallySpeaking for its friendlier training, speed, ability to add vocabulary easily, and overall better accuracy. If you're not sure whether speech recognition is for you, then ViaVoice is an inexpensive way to get started, plus it integrates nicely with Word and WordPro. But for serious productivity, catch hold of the Dragon's tail.
Product Information
Natura
llySpeaking Personal Edition...........................$150.00 (street price)
(133-MHz or faster Pentium, 32/48 MB of RAM [Windows
95/NT], Sound Blaster 16-compatible sound card,
60 MB of disk space)
Dragon Systems, Inc.
Newton, MA
Phone: 617-965-5200
Internet:
http://www.dragonsys.com/
Enter HotBYTEs 1033.
Information on
this product
and similar products (software applications)
Do You Hear What I Say?
