Speaking Solutions | 1-888-749-1844

Speech

The Dragon Scale

By Karl Barksdale
Speaking Solutions © 2003, 2004, 2005

Dragon NaturallySpeaking is the standard against which all other speech recognition programs are judged. In this article we will rank the various products from 1997 to the present against versions of Dragon 3.0 to 8.0. This is a semi-chronological listing that provides a useful way to track the improvement of speech recognition products against Dragon over time.

Dragon NaturallySpeaking 8.0 (2004-2006)
Microsoft Windows Vista Speech Recognition (2006) Not yet Available

DRAGON NATURALLYSPEAKING 8: When ScanSoft claimed that version 8 is 25% more accurate than version 7.3, I was skeptical. So I loaded the software and trained it. It's better. Noticeably better. I don't know how to judge if it is 25% better, but I like the accuracy. After all, accuracy is the thing that matters most with speech recognition.

The new natural commands make version 8 the best version of Dragon ever. For this reason, we felt like we needed to drastically upgrade our Nifty 58 training manual. With so many new commands available in Microsoft Word and Excel, it would seem a shame to offer the same old Dragon instruction.

MICROSOFT WINDOWS VISTA SPEECH RECOGNITION: The last three versions of Dragon NaturallySpeaking (6, 7, and 8) stretching over four years, were all clearly better than the competition. In fact, there was very little competition for Dragon. However, we were able to try Microsoft's new Windows Vista speech recognition while still in beta. And it was impressive -- very impressive. And while we will have to wait until the final version hits the market to be sure where to accurately place this product on the Dragon scale, Vista SR looks promising. It's been almost 5 years since there was any competition for Dragon, and it will be nice to see a horse race once again. (Read our full review of Windows Vista SR's Beta)

Personally, I can't wait to see what the next generation of products has in store for us.

Dragon NaturallySpeaking 7.0 (2003)

Improved accuracy and even more attention to Natural Language Commands in Microsoft Office products has given version 7 a clear lead. Combined with price decreases for school licenses, Dragon NaturallySpeaking is currently at the top of its game. Initial enrollment time has now fallen below five minutes for those with a fast PC. Despite shorter enrollment times, accuracy is now between 96-98% for trained users with clear enunciation. The error correction system was also improved again.

Dragon NaturallySpeaking 6.0 (2002)
Microsoft Tablet PC Speech-recognition (2005)
IBM ViaVoice 10 for PCs (2002)

New error correction systems and improved command and control (also called Natural Language Commands) gave these products a distinct advantage over everything else on the market at that time. Initial enrollment times plummeted to under 10 minutes. While they could still run with 128MB of RAM, 256MB of RAM and Pentium III 800MHz or faster processors were becoming the minimum. Accuracy rates improved to 95-97% after initial enrollments for trained users with clear speaking skills.

Dragon NaturallySpeaking 5.0 (~2000)
L&H Voice Xpress 5.0 (~2000)
IBM ViaVoice Millennium (PC ~2000)
IBM ViaVoice Macintosh Enhanced/OS X
Microsoft Tablet PC Speech-recognition (2002)
Microsoft Office 2003 Speech-recognition (2003)

These were the first products that could truly be used by a large audience of users. No longer limited to those with repetitive strain injuries, initial training could be accomplished in under 15 minutes and would achieve remarkable initial accuracy over the 95% threshold! Albeit,128MB of RAM became essential. Pentium III 500MHz processors were the new minimum. The command and control systems became much better, particularly with Voice Xpress and Microsoft Tablet PC Speech Recognition. Accuracy and error correction systems improved even further in Dragon NaturallySpeaking, Office 2003, and IBM ViaVoice.

Dragon NaturallySpeaking 4.0 (~1999)
IBM ViaVoice PC (pre-millennium editions 1998 to 1999)
IBM ViaVoice Millenium Edition for Macintosh (2000)
L&H Voice Xpress 4.0 (~1999)
Microsoft Office XP Speech Recognition (2001)

These products truly foreshadowed the future. Dragon continued to be the accuracy leader. But IBM took a chance on Macintosh, a chance that paid off, making ViaVoice the uncontested leader in Macintosh speech programs. Enrollment times dropped to approximately 20 minutes. Initial accuracy improved to the 92-94% range. However, memory requirements expanded to 96MB-128MB for comfortable dictation.

Dragon NaturallySpeaking 3 x. (1997)
IBM Simply Speaking Gold (1997)

These were the first continuous speech recognition programs that actually worked! THE pioneering speech programs, they were the result of decades of IBM and Dragon research. If you were suffering with a repetitive strain injury, and were willing to read a minimum of 45 minutes, you could achieve a remarkable 90-92% accuracy on Pentium II 266MHz computers with 64MB RAM with careful enunciation.

(Disclaimer: The main purpose of this article is to chronicle a history of the progression of speech-recognition improvements over time. Unfortunately, because of limited resources and time, the list is not comprehensive and inclusive of every release of every speech recognition product. The comparisons for this editorial are our estimates based on the initial reading of a training script. Estimates are based upon my opinion and experience. Initial accuracy will vary from user to user depending on enunciation skills, level of training, and the quality of their headsets and sound cards. Accuracy of any product can be improved by reading additional training stories and correcting errors in the proper way. Truth is, if you work hard enough with any one of these products, it will probably work well for you.)