Law Practice Today

Introduction – Voice Recognition: Does it Work?

A lawyer who talks to herself is crazy, right? Crazy like a fox! Speech recognition software has arrived - and now it's for more than just creating documents. Today, smart lawyers are using digital dictation devices as well as voice recognition software to complete and manage a myriad of computerized tasks – often without support staff. Read on to learn how you can implement and use voice recognition software, and how you can pair it with the latest in digital dictation equipment to improve your efficiency and redirect staff efforts – or even eliminate those costs entirely.

At TECHSHOW 2002, Wells Anderson and David Hirsch did a presentation entitled “Speech Recognition: Real Uses in Real Life.” The first heading in Well’s handout was “Does it Work?” We both attended that presentation and we remember that Wells answer was a hearty “It depends.”

A lot has changed in five years, but I think our answer to that question today is still a hearty “it depends.” There are a lot of factors that go into how well speech recognition software works, and we’ll get to them. But first, a quick detour to let you know what is available now in the speech recognition field and what has changed in the last five years.

Who’s In the Market, and Who’s Not

In 2002, the last time that a TECHSHOW session compared the available options for voice recognition in the law office, the program was a sort of “shoot off” between IBM Via Voice and Dragon Naturally Speaking, the only real players in the voice recognition game at the time. The landscape looks a little different today, with mergers, acquisitions and a disappearance of a sort. And, there’s a(n old) new kid on the block, too.

IBM Via Voice

IBM still engages in direct sales of something called Embedded Via Voice, but it isn’t the Via Voice lawyers are probably used to. According to the IBM website, Embedded Via Voice is described as “mobile and speech middleware” and is aimed at “B2E/B2C services, including voice recognition and telephony speech processing.” In other words, it is not being developed for use in cars, cell phones, ovens, refrigerators and other things that will eventually be able to be operated by voice commands.

A Via Voice consumer product is still available, although not directly from IBM. If you’re interested in Via Voice, you can find it at lots of on-line retail software sites. All you have to do is put the phrase IBM “via voice” into Google, and lots of places where you can buy it will come up. I found several different versions, including IBM Via Voice for Windows Advanced Version 10.0 for about $80, and Via Voice for Mac OS/X Edition for about $130.

If you shop for Via Voice online you should carefully review the product before you buy. For example, we found something for sale called Via Voice PRO USB for $164, which sounded quite good until we read over Wells Anderson’s old handout. VV PRO USB was being offered back in 2002. When we carefully read the information on PRO USB, we discovered that it will work with previous versions of Microsoft Office, but won’t work with Office 2003.

Dragon Naturally Speaking

We can’t say whether Via Voice was not as good as Dragon, or if it just lost the race – much as Beta did to the VHS format. Either way, though, Dragon seems to have developed the larger market share and has become the current standard for consumer voice recognition.

Since we last discussed Dragon Naturally Speaking at TECHSHOW, the product has been bought by Nuance Communications, Inc. (www.nuance.com) and the most recent version is Naturally Speaking 9. After a quick detour, the remainder of this paper will concentrate on that product, because that’s what both your authors currently happen to have and use.

Microsoft

When it comes to software, at least on the personal computer, one should never count Microsoft out.

Beginning with the operating system Windows® XP, Microsoft has included some voice recognition features, as well as text to speech (let your computer talk to you), as a part of the operating system’s features.

In order to use these features you'll have to have the speech recognition engine loaded as a part of your operating system. To check to see if it was loaded when the operating system was installed, you will need to click Start, go to the Control Panel, and select Speech. You will see a Speech Properties dialog box. If the voice recognition engine is loaded you will see tabs for both Speech Recognition and Text-To-Speech. If the voice recognition engine is not loaded, you will only have the Text-To-Speech tab. If the voice recognition engine is loaded, you should be able to plug-in a microphone, take appropriate levels, and attempt to dictate into any of the Microsoft Office XP components.

According to the Microsoft web site, these features will be greatly enhanced in Vista, their next generation operating system, allowing users to dictate documents, use voice commands to start and switch between programs, control the operating system and fill in forms on the Internet.

This may be a way for those who are interested in speech to text, but are not ready to purchase a speech recognition program, to get a taste of what it’s like.

(Note: Around press time [ed.: in January 2007] we talked with an industry insider who emphatically said that the speech recognition package included with Microsoft Vista is surprisingly robust. We cannot testify to that as neither of us has a computer with Vista installed.)

Dragon Naturally Speaking 9: How Does it Work?

Versions Available: Which one is right for you? Dragon NaturallySpeaking Version 9 comes in four different versions: Standard, Preferred, Professional and Legal.

• Standard - We do not recommend that lawyers invest in the Standard Edition. It retails for $99, but it is not as feature rich as Preferred.
• Preferred - For most lawyers, we recommend the purchase of Dragon Preferred Edition 9. In fact, we feel that this product, which retails for $199, is a real bargain. It has the basic voice commands and generally does a very good job. Because of the significant price difference between this product and the two next versions, we believe that most lawyers are better served to start with this and then upgrade later if they feel an upgrade is warranted.
• Professional and Legal - You generally have to get a quote from a dealer for both of these products, which retail in the $800-$1000 range.

It is our understanding that the voice recognition capabilities of these products are essentially the same as with Preferred. They do have additional features that some lawyers will find valuable. For example, both have the ability for complete audio replay of a dictated document. This allows a third party to proof your work.

Because Dragon’s errors are inserting the wrong words that it thinks it hears, rather than misspellings or typos, it is very difficult for anyone other than the person who originally did the dictation to proof the document. Preferred will allow one to replay the sound files of paragraphs that were dictated and listen to them, but it takes the more expensive editions to replay the sound file from an entire document. So if you want your secretary to be able to proof your dictated work in Dragon, you will need to invest in two copies of the more expensive version: one for you and one for him/her.

The Legal version is also preconfigured to allow proper recognition of legal citations and it has a dictionary of frequently used legal terms. Some lawyers have told us that these are "must have" features. Because neither of us use many legal citations in our work, this isn’t the case for us. Most of the Dragon users that we know use the Preferred edition.

What else do I need to make it work?

You will need a relatively new computer if you want Dragon NaturallySpeaking 9 to function at the level that it was intended to and is capable of.

Minimum system requirements for a PC include a Pentium 4.1 GHz processor (or equivalent AMD), 512 MB RAM, 1 GB free hard drive space, Windows XP (SP 1 or higher) or Windows 2000 (SP 4 or higher), and a Creative® SoundBlaster® or equivalent sound card supporting 16-bit recording.

As a sort of "he said/she said" Jim found that the system really didn't work very well with less than a full gigabyte of RAM. Laura, on the other hand, found that the system worked fine on her laptop with 512 MB RAM, and couldn't tell a real difference when she upgraded to 1 GB of RAM.

As another "he said/she said" Jim swears by his USB microphone, but Laura has been perfectly satisfied and has had good results with the microphone that came with Dragon NaturallySpeaking 9 Preferred.

Installation: How hard can this be?

Installing Dragon NaturallySpeaking 9 is so easy there's hardly anything to say about it. You just put the disk in your CD-ROM drive and follow the instructions.

Training: I haven’t got all day!

Once you have installed Dragon NaturallySpeaking 9, you will be given the option to train it. Training is no longer mandatory, but we recommend that you go through the process anyway. You will be given several options of things to read into the program, and the program will almost always complete the training before you have gotten more than about five minutes into the reading.

The program will also give you the option of reviewing your writing style by scanning your e-mail and other files on your computer. This process will take less than 20 minutes, and will greatly improve the program's accuracy right from the start.

All in all, you can do enough training in about 30 minutes to really improve your initial satisfaction with using voice recognition.

Recognition Effectiveness: Pardon?

Dragon NaturallySpeaking 9's ability to convert your speech into text will have a lot to do with your accent. If you are from the Midwest, you will probably be going like greased lightning right out of the box. If you have a strong accent from, say, the Bronx, Boston, or the South, you can expect to have a little more trouble and do a little more training. In the end, it will be worth it.

Gadgets: What else can I use with it?

Some lawyers have reported excellent results using handheld digital equipment and then feeding that dictation into Dragon for processing. Neither of us has tried that at press time and, unfortunately, we could not find a list of certified devices on the Nuance website.

While we hate to single out a single vendor, we would be remiss in our duties if we did not point out the website www.knowbrainer.com as a source of information and related products.

Jim ‘n’ Laura’s Top Ten Voice Recognition Tips

1. Read the User’s Guide first.

If you’re like most lawyers, you know you’re smart and the installation and training is so easy that you’ll be lulled into thinking that you don’t need to look at the User’s Guide. You can probably get by, but the problem with this approach is that you’ll develop some bad habits that you could just as easily do without, not to mention that you’ll keep your own blood pressure lower by being aware of all the special “commands” that allow you to edit and format what you dictate.

2. Use a USB Microphone.

We know we are repeating ourselves, but this is an important point. Use a USB microphone, not the standard one that plugs into your sound card. While it is not necessary, in our experience, to buy an expensive microphone, it is necessary to buy one that is configured for speech recognition and has noise canceling features.

3. Get Plenty of Memory.

We believe that 1 GB of memory will work for most users. But more is always better. 2 GB will probably mean great speech recognition results and your other applications will run much faster as well.

4. Speak in Phrases.

I think of my Dragon as sort of like a dog. He probably couldn’t understand the Gettysburg address, but he’s really good with short phrases that he hears over and over again – especially if they’re said with the same inflection each time.

Most of us don’t speak in perfect, fully-formed sentences nor do we speak one – word – at – a – time. Instead, we speak in idea couplets, or phrases. If you will dictate in natural word groups that flow, Dragon will almost always get it right. And, just as you wouldn’t yell at your beloved puppy, don’t yell at Dragon, either!

5. Get a single headset that can be used with both your dictation program and your telephone.

I adopted a headset for my telephone, reluctantly, when I started having neck problems from cradling the phone on my shoulder while trying to talk and use the computer at the same time. I couldn’t get the phrase, “Would you like fries with that?” out of my mind, but I can speak and hear clearly on the phone while having full access to my computer and practically everything else on and around my desk. Because I had just gotten and paid for the phone headset when I got the voice recognition software, I wasn’t in a position to ask for another “special” headset that would work with both computer and the phone.

Although a perfectly adequate headset comes with Dragon Naturally Speaking 9 Preferred, shop around and get one with a switcher that will also allow you to answer the phone. As you dictate more and more each day, you’ll find it a big relief not to be constantly switching from one headset to the other.

6. Take care to avoid sending e-mails by accident.

When the voice recognition software makes errors, they are almost always correctable. One exception to that rule is with the e-mails. You can dictate e-mails just as you would dictate documents in most e-mail applications. (Gmail has not worked well for this purpose for some reason.) But Dragon also recognizes voice commands in e-mail such as "send." The problem can occur when Dragon misunderstands a word as the command “send” and sends the e-mail before you're ready. This can happen more frequently than you think; for example, Dragon may have misunderstood the word “then” as “the.” When you select “the” and say “then” as a single word, the odds are pretty good it will interpret that as a command since it is a word instead of a phrase. And the most likely command is “Send.”

We therefore recommend that you do not fill in the To: field or other address lines and e-mail until it is completed and finalized. If you are doing a reply, then we suggest that you highlight all of the addresses in the To: field and use Control + X to cut them out and copy them to your clipboard. You can paste them back in after you have finished dictating and proofing the e-mail.

7. Correct more than one word at a time for best results.

Because Dragon 9 decides what word you are saying in the context of all the other words around it, your best training opportunity will sometimes come from correcting more than one word at a time. For example, if you keep saying "in the wash" and Dragon keeps typing "in the watch" you're better off to correct the entire phrase than to keep working with the single word.

8. Check your microphone before every use.

We all want to get started as quickly as possible, and there's a tendency to try to get things set up and then leave them as they are. If it ain't broke, don't fix it. You will have better results dictating with Dragon NaturallySpeaking if you check your microphone's position and volume level before each use.

9. Know where your speech file is, and make sure it’s part of your backup protocol.

There is nothing more aggravating than to spend weeks using the system and letting it get to know the sound of your voice better than your mother does, only to lose the file when you upgrade computers, get a new hard drive, or have a disc crash. In Windows 2000/XP (both Home and Pro)/Server 2003, Naturally Speaking 9 files will be found in the following locations:

C:\Program Files\Nuance\NaturallySpeaking9\

C:\Documents and Settings\All Users\Application Data\Nuance\NaturallySpeaking9 C:\Documents and Settings\<username>\Application Data\Nucance\NaturallySpeaking9

10. You will probably be most efficient by continuing to use your mouse and/or keyboard on occasion in conjunction with Dragon.

Clearly, it is important to use the software rather than the keyboard to correct words when Dragon has misunderstood you. This correcting process allows the software to learn and improve its recognition capabilities. However, there are still some times that the mouse can be used to quickly perform an operation or part of an operation much more efficiently than you can convince Dragon to do by voice.

Suppose, for example, you dictated a 10 word book title. Dragon didn't recognize that the title needed to be capitalized and italicized. It is probably easier to select the entire book title with your mouse and then use the voice command “cap that” and then to “italicize that” than to try to capitalize the entire title with one operation. It would be very time consuming to select and capitalize each word using voice commands and getting Dragon to select a phrase, unless it was dictated as a phrase, is a bit tricky.

Sometimes as you review your work, you will find that you want to change, rather than correct, a word that is contained in a sentence you dictated several sentences ago. In that case, it is often better just to select that word by double-clicking on it then select it with the voice commands, as you may have dictated that word or word that sounds like it in a subsequent sentence. Dragon will first select the most recent use of a word.

The point is still to get the work done, not to train the software. After a couple of unsuccessful tries at correcting a word, it is probably best to just select it with your mouse, correct it with your keyboard and move on. But you should only use this technique sparingly, since voice corrections are the only way to improve your speech recognition data file.

Conclusion

Whether speech recognition products make sense for you in the year 2007 is dependent on several factors. Are you willing to invest the time to train the product? Are you motivated to truly use the product? Perhaps the biggest variable is your present typing speed.

We have heard of lawyers who have had physical challenges that prevented them from effectively using the keyboard but who were very efficient with using prior versions of Dragon.

For most of us, version 9 Preferred was the point at which the product finally was usable to produce work, instead of just being an interesting technological curiosity.

If you're a very slow typist who needs to use the computer to produce documents, we suggest that you immediately purchase Dragon 9 Preferred Edition. If you are a typist who can produce work at 80 to 100 wpm, this product is probably not for you.

This product is also recommended for lawyers who are spending so much time keyboarding that they are physically tired or stressed from doing so. Repetitive stress injuries from keyboarding are becoming more and more common in the workplace. Learning to talk to yourself, or your computer, may be the key for many people to avoid pain, and sometimes injury.

Jim Calloway is the Director of the Oklahoma Bar Association Management Assistance Program. He publishes the blog Jim Calloway's Law Practice Tips and was the co-author of the book Winning Alternatives to the Billable Hour (2nd ed.) (2002 ABA.) He was chair of ABA TECHSHOW 2005. He is a member of the Council of the ABA's Law Practice Management Section and chair of its Practice Management Advisors Committee. He is co-chair of the 2007 GP/SOLO National Solo and Small Firm Conference. He has been named a Fellow of the College of Law Practice Management. He writes Law Practice Tips column for the Oklahoma Bar Journal. The archives of that column are available online here.

Laura A. Calloway has served as the Director of the Alabama State Bar’s Practice Management Assistance Program since 1997. Before joining the Alabama State Bar to establish the program, she practiced law in Montgomery, Alabama, for 16 years as a solo practitioner and member of two small firms, concentrating her practice in residential real estate, collections, consumer bankruptcy and family law. She is a member of the ABA TECHSHOW 2008 Planning Board, and is a regular contributor to Law Practice Magazine.