NatSpeak Version 4.0

Last Modified: November 23, 1999

On Aug. 4, 1999 Dragon Systems announced version 4.0 of the Dragon NaturallySpeaking family. Instead of simply rehashing the press release, I wanted to give you some more personal opinions on what is really new in Dragon NaturallySpeaking 4.0.

Here is a short list of the most significant features of Dragon NaturallySpeaking version 4.0:

• Better accuracy
• Shorter training
• Incremental Vocabulary Builder
• Web browsing by voice
• Teens models

Better Accuracy

The most significant new feature of Dragon NaturallySpeaking 4.0 (in my opinion) is improved accuracy. High accuracy is the number one feature requested by users of speech recognition programs, and Dragon Systems has worked hard to improve the accuracy of Dragon NaturallySpeaking in this new version. There are a number of different modifications to the basic algorithms and speech models which contribute to this improved accuracy.

As with each upgrade of Dragon NaturallySpeaking, in order to get the full effect of the improved accuracy you will have to retrain. You should also rebuild your vocabularies starting with new base vocabularies. You will get some minor accuracy improvements without retraining and rebuilding the vocabularies, but it will not be as noticeable.

The new vocabularies in Dragon NaturallySpeaking have significantly more words active. By having more words active, errors caused by words being out of vocabulary are significantly reduced. The actual number of active vocabulary words is quoted in the press release as 160,000, but this is an effective number. If you bring up the Vocabulary Editor, you will not see a list of 160,000 words because of the way the vocabulary size was increased.

The acoustic models have also been redesigned to produce higher accuracy dictation. And, there have been other improvements as well, including some improvements in noise robustness and some modifications which will increase accuracy when using the Dragon NaturallyMobile pocket recorder.

Most of these advantages, however, are only available if you create a new user since if you use your existing user you will not get the modifications made to the vocabulary and acoustic models. (This was a similar issue when version 3.0 was first introduced with BestMatch technology.) Which brings us to the next feature:

Shorter Training

Dragon NaturallySpeaking 4.0 includes the new BestMatch III models which were introduced earlier this year. BestMatch III is designed to be used on higher end computers. I understand that the minimum requirement for using BestMatch III is a Pentium II 300 MHz with at least 64 MB of RAM. But I would recommend 128 MB of RAM myself.

With BestMatch III, training time is reduced from 18 minutes of dictated text to three minutes of dictated text. And the accuracy after just three minutes of enrolling, is more better than previous versions of Dragon NaturallySpeaking using 18 minutes of enrollment time. This makes it very reasonable to consider installing Dragon NaturallySpeaking version 4.0 and creating a new user from scratch.

Recently, I installed Dragon NaturallySpeaking 4.0 on my machine at work. After the initial three minutes of training, my accuracy dictating into Dragon NaturallySpeaking version 4.0 was as high as my highly tuned models created using Dragon NaturallySpeaking version 3.52 and optimized with months of consistant error correction. Of course, to get the full benefit you will want to create a new vocabulary which means that you will have to run the Vocabulary Builder again. Which brings us to the next feature:

Incremental Vocabulary Builder

As previously documented on this web site (see Vocabulary Builder), in earlier versions of Dragon NaturallySpeaking every time you run the Vocabulary Builder, it rebuilds the statistics of word usage based only on the documents used in that execution of the Vocabulary Builder. This makes it difficult to incrementally improve your vocabulary since you are forced to maintain your own list of source files so that you can rerun those files through Vocabulary Builder when you need to add more files.

But now with Dragon NaturallySpeaking version 4.0, the Vocabulary Builder automatically remembers the statistical information every time it is run. The second and subsequent time you run the Vocabulary Builder, any documents you specify are added to the original set automatically. This makes it possible to add one or two documents to your vocabulary models by running the Vocabulary Builder incrementally.

Web Browsing by Voice

New in Dragon NaturallySpeaking version 4.0 is support for running Internet Explorer by voice. This feature allows you to verbally control the browser by speaking the names of the links. You can also move to and dictate into text fields in forms.

Warning strong personal opinions follow:

Web browsing by voice is an often requested feature, and a couple of other speech recognition companies offer similar implementations. The implementation of web browsing by voice in Dragon NaturallySpeaking is very good, but that does not make the feature usable. It is my personal opinion that web browsing is not an activity which is conducive to use by voice. Web pages are not designed to be conveniently controlled by voice, and the basic browsing activity is very visual not verbal. It is also difficult to speak web addresses (even though Dragon NaturallySpeaking has support for automatically formatting them).

That said, I do not want to detract from this potentially useful feature, especially for those users of Dragon NaturallySpeaking for using the product to avoid having to use their hands for one reason or another. But for those of us who are comfortable using the mouse, I do not believe that this feature is significantly useful in the long-term.

Teens Models

Finally, Dragon NaturallySpeaking version 4.0 includes the same team voice models which were introduced in Dragon NaturallySpeaking for Teens. This makes it reasonable to consider buying a copy of Dragon NaturallySpeaking version 4.0 for the entire family since it now contains both adult and teens models.

In Conclusion

The press release for Dragon NaturallySpeaking version 4.0 lists a number of other features. Some of those are simply features of the Dragon NaturallySpeaking family and have not changed (significantly) in version 4.0. Some of the other features which are listed in the press release (like the New Setup Wizard) are minor and not worth going into details about.

However, I believe that current users of Dragon NaturallySpeaking should seriously consider upgrading to version 4.0 if for no other reason than to recognize the higher accuracy in this new version.

Warning: if Dragon remains true to its historical behavior (and I have no reason to believe that Dragon has changed), then you will probably see version 4.0 first appear in retail, with professional editions coming later, and upgrades for existing customers coming last. I wish it were otherwise, but Dragon is still a relatively small company and there are now dozens of different variants of Dragon NaturallySpeaking which have to be builtand tested every time a new version is released.

P.S. NatLink (see Python Macro System) works fine with version 4.0 although I have not tested my other utilities.

This web page (http://www.synapseadaptive.com/joel/NatSpeak40.html) was last modified on November 23, 1999. The contents of this page are (c) Copyright 1998-1999 by Joel Gould. All Rights Reserved. See Copyright Information for more details.