Voice Dictation text

Voice Dictation

Overview

I have a client who sent me an e-mail enthusing about his new voice dictation software - Dragon Systems NaturallySpeaking Preferred. He was so enthused he ordered a new computer to replace a not quite 2 year old, but quite respectable, DELL Pentium 166. His new machine is a Pentium II 400 MHz, 256 MB RAM, with a 7200 rpm SCSI Ultra/Wide hard drive. (We had to draw the line somewhere!)

I have been using NaturallySpeaking Preferred myself since receiving his e-mail. It is awesome, even on my puny Pentium II 266 with only 96 MB RAM and a slow (in comparison) IDE Ultra/DMA hard drive. I've learned it doesn't take a muscle machine, if you're willing to compromise a little.

The Best

But I digress. The result of my testing both NaturallySpeaking Preferred and Lernout & Hauspie's VoiceExpress Pro was to discover that Dragon's product is significantly superior. It is the one I use. I get virtually 100% accuracy when I speak clearly. The occasional correction is painless and psychically rewarding. That is, correction can be done by voice, and the intangible but positive feedback from the correction process makes you feel as if the recognition is getting better. And it does improve with each correction.

Second Best

I was very disappointed by the Lernout & Hauspie program. I never had a feeling with L&H that it was getting better so I dropped it. Particularly after reading a review by an InfoWorld columnist who thought it was far better than the Dragon system. Not from what I saw it's not. And this is the Microsoft "house brand", and I'm a Microsoft bigot. Maybe the next version of L&H will be better. Or maybe it will approach Dragon by the time it is embedded in Windows NT 5.5 (or 6.0). I haven't uninstalled it yet, but I don't bother with it anymore.

Compromises

My client was quite correct to specify his new workstation with a SCSI hard drive and extra quantities of RAM. This is typically how I build a server. The difference between his SCSI and my IDE (Ultra/DMA!!!) is significant. It takes a full minute after I click on the "Use NaturalWord" menu item before I can start dictating. The 80 second delay is spent thrashing the hard drive. It's official! Ultra/Wide SCSI beats Ultra/DMA IDE by six to one. I haven't played enough with his system to know for sure, but the one test I ran had NaturallySpeaking listening within 10 seconds.

How much of a compromise is this? Once the Dragon system is initialized I can switch around my typical applications and then come back to Word97 and begin dictating within 15 seconds. This is where having gobs more RAM would help.

Negative Effects

NaturallySpeaking is not fool proof. You better know your way around a computer and not be terrified if you have to re-boot. Sometimes Word97 kind of "locks up" or NaturallySpeaking doesn't respond quite right - just re-boot. And, now, I always get a Dr. Watson error when I exit Word97. A major pain to be sure.

Rewards

Almost 100% accurate recognition of continuous speech dictation.

Applications

The voice dictation technology can't be applied everywhere. It is really only appropriate, currently, for writing chores. Except for the degraded interface, this is my preferred method of creating e-mail. I say degraded interface, because although I am using Word97 as my e-mail editor in Outlook 2000, NaturallySpeaking doesn't perform the same. The most annoying aspect is not having the smart space function of Word97. This is when you change a word or phrase and Word97 knows how to space around the change. Evidently, when Word97 is used from Outlook 2000, the smarts are turned off. (Kind of like the idiots who designed the FrontPage98 page editor - it shares nothing in common with Word97. Why?)

Requirements

I think my system is adequate: Pentium II 266, 220 MB RAM, 4 GB Ultra/DMA IDE. Here is what I think you really need:

The fastest math co-processor in existence, the Intel Pentium III. The current generation of chip is the "Coppermine" series. Specifically, you want the 133 MHz Front Side Bus ("B") where the L2 Cache runs at the processor speed ("E"). I think this is the 633 MHz version, but the 533 MHz is certainly adequate.
At least 128 MB RAM. The "Best Match" technology in NaturallySpeaking v3.01 requires more than 64 MB under Windows NT. You are running Windows NT on all your network workstations aren't you? If not (???) CALL ME - WE NEED TO TALK.
At least a 4 GB Ultra/Wide SCSI Hard Drive.