After six tries, Dragon NaturallySpeaking may have finally gotten it right. PC speech recognition has been around for awhile. And successive editions were increasingly accurate, but the program always ran so sluggishly that it couldn't compete with most competent typists.
But that's changed with Version 6, which merges the best of Dragon NaturallySpeaking 5 and L&H VoiceXpress 5 -- and which is now sold by ScanSoft, the OmniPage Pro, TextBridge Pro, and PaperPort OCR vendor that's acquired the bankrupt Lernout & Hauspie's speech-related assets, including those L&H got when it acquired Dragon Systems. Expect ScanSoft to emphasize the Dragon brand over the L&H label.
NaturallySpeaking 6 is fast and responsive; it reacts crisply and quickly to both voice commands and dictation. Or at least it does on a modern PC like my 1.9GHz Pentium 4 desktop with 256MB of RDRAM -- which far exceeds ScanSoft's optimum recommendation of a 500MHz Pentium III or compatible with 256MB, which in turn seem a lot more realistic than the advertised minimum of a 400MHz Pentium II with 182MB, along with a Creative Sound Blaster 16 or compatible audio card capable of 16-bit recording.
But there's clearly more than just muscle at work here -- Microsoft Office XP includes a speech recognition feature, but its performance on the same PC pales in comparison to the new NaturallySpeaking's.
On the other hand, so does its price -- free with Office XP, while the "Professional Solutions" edition of NaturallySpeaking 6 tested here costs $695. It's the ultimate business-oriented version of the software, a counterpart to a $795 edition for public safety workers and $995 (apiece) releases for the legal and medical professions.
Most PC users will probably opt for the $199 Dragon NaturallySpeaking 6 Preferred, a mainstream version lacking some Solutions Series features such as the ability to record speech macros and save your dictation for review by a transcriptionist; sophisticated step-by-step scripting compatible with Microsoft Visual Basic for Applications; and natural-language commands for Microsoft Outlook and PowerPoint and Lotus Notes.
The Preferred version joins its deluxe siblings in supporting Microsoft Word, Excel, and Internet Explorer, Corel WordPerfect, and America Online. If you remember the even simpler and cheaper NaturallySpeaking Standard and Essentials packages, they're still at version 5.
Speak Up and Running
In addition to a tremendous boost in performance, version 6 boasts a streamlined setup routine. It took about 13 minutes to install the software and another 13 minutes to train it to recognize our voice; previous incarnations took as long as 45 minutes for the latter.
NaturallySpeaking 6 also brings a boxcar of new features, including VoiceXpress 5's "Nothing But Speech" filter that ignores "um" and "ah" pauses as well as ambient sounds like breathing. The filter isn't perfect, but it's a noticeable improvement over past versions.
The program offers several modes -- command, dictation, spelling, and numbers -- which you choose by simply saying, for example, "Switch to dictation." The modes can improve recognition speed and accuracy for tasks like dictating numbers into an Excel worksheet, but we found NaturallySpeaking surprisingly smart about interpreting what we said -- whether dictating text or giving program commands or navigational instructions -- without deploying specialized modes.
Another feature to increase accuracy is the Acoustic Optimizer. As you dictate text and correct your mistakes, NaturallySpeaking stores information about your speech patterns which it can use to improve its accuracy. This configuration personalization can take hours even on a speedy PC, however, so you should start it when you've finished your work for the day.
Making corrections, which was formerly tedious and vexing, is also much improved. Misspell mangement (sic)? Simply say, "Select mangement," repeat the word, and the correct spelling appears. Another significant convenience is the ability to dictate individual keystrokes -- you can say, "Press Delete" or "Press Backspace," and Dragon will do your bidding.
This version of NaturallySpeaking also has powerful e-mail features. The program isn't flustered when you dictate an e-mail address; in fact, you can add addresses to its vocabulary. And within e-mail programs such as Outlook, AOL versions 5 and 6, and Eudora, you can execute speech commands for opening, closing, deleting, creating, sending, and replying to messages.
Dragon's adaptation of speech commands to Microsoft's Web browser is particularly impressive. Not only can you perform perfunctory Internet Explorer tasks such as "back" and "forward," but you can recite linked words on a Web page and "click" on them with your voice.
As mentioned above, NaturallySpeaking 6 Professional will record what you say so you (or someone else) can play back your dictation later, resolving any confusion over exactly what you said. The software even lets you save your dictation along with your document, or dictate while away from your PC into a number of digital tape recorders.
Not only will the program play back your dictation, but it will read documents to you, too. Unfortunately, although L&H's RealSpeak is touted as the most human-sounding text-to-speech voice to date, it sounded painfully synthetic and grating in our tests.
Speech recognition software was introduced years ago with sky-high promise, but faded from view due to erratic performance that made PC dictation more frustrating than fruitful. With the impressively improved NaturallySpeaking 6, the promise of speech input may finally be realized.
Manufacturer: Scansoft, www.lhsl.com
Pros: Good speed; helpful speech commands; e-mail support.
Cons: Synthetic text to speech.