ViaVoice – Review

ViaVoice
Company: IBM
Estimated Price: $89.95
http://www-4.ibm.com/software/speech/mac

I’m going to start this review by speaking to ViaVoice. If I’m still using it when I’m done, how bad can it be?

After several years without a current dictation product, three different companies have recently promised to release their products for the Mac. IBM is the first one to actually ship a product, as they released ViaVoice Millennium Edition just in time for the holidays. Although it is exciting to have such a product available for the Mac, ViaVoice has enough flaws and quirks to make using it a bit frustrating.

Who needs ViaVoice?

Before investing in ViaVoice, you should decide whether you need a dictation product at all. Those people with well developed typing skills should be able to outperform the speed at which this program recognizes their words, while users with disabilities or prone to repetitive stress disorders however, may find ViaVoice invaluable. One should also consider the environment in which it will be used. The people who share your home or office may not appreciate having to listen to you talk to your computer all the time. In addition, ViaVoice has steep hardware requirements, requiring a computer that shipped with a G3 processor at the minimum. I’ve tried it on both a 266 MHz and a 450 MHz G3 and performance on the slower machines was barely acceptable. On the speedier equipment, the program performed much better, although the added processing power did not improve its accuracy.

Installation and set up

IBM includes a small manual that thoroughly covers all the softwares functions. Despite its underlying complexity, ViaVoice has a fairly simple interface, and a small manual seems to be enough. In addition, a reference card is included for quick access to all the commands and punctuation macros that you can use. The package also includes a superb microphone headset that includes interchangeable colored plastic parts that let you customize it to match your iMac (excepting graphite, at the moment).

ViaVoice installs quickly and painlessly from a CD. Although the installer places a large number of files and libraries on your hard drive, almost all of them are put in a single folder, cutting down on hard drive clutter. All the features of the software users need access to are available through a menu extension that’s installed at the same time. From the menu, you can activate a set-up assistant to get you started with voice recognition. The assistant will help you configure the microphone to get the proper level of sound input. After that, you are prompted to read any or all of a number of passages to help the software get accustomed to the sound of your voice. Once finished, ViaVoice will analyze what it has just heard. This process immobilizes your computer for a significant amount of time and results in the creation of what IBM terms a “voice model.” When it’s done, you’re ready to begin dictating.

The accuracy of the voice model determines how well ViaVoice recognizes what you speak. Fortunately, the model can be improved. From the ViaVoice menu, you can choose to read further passages at any time. Although it comes with a huge vocabulary of words, names, places, and technical terms, ViaVoice can analyze text documents you’ve selected for words it does not recognize. You’ll then be asked to pronounce those words so that they can be added to your voice model. Finally, ViaVoice attempts to improve the voice model each time you make a correction to previously dictated text. The software supports voice models for multiple users, and allows voice models to be exported and imported for use on other computers.

Using ViaVoice

ViaVoice does not allow you to dictate into just any text editing application. Instead, you dictate into its own text editor, called SpeakPad, which only allows the most basic formatting information to be entered. Since you can’t do everything you’d like in SpeakPad, IBM has included commands that will allow you to transfer the text from it to any of five programs, including Microsoft Word, Netscape Communicator, and Outlook Express. To use any other programs, you have to transfer the text to the clipboard and then paste it into your document manually.

In addition to recognizing text, ViaVoice can recognize a variety of commands and macros. Commands allow you to enter formatting information, move the cursor, and correct text. Speech macros provide the program with a great deal of flexibility. Macros range from the simple ones included, which handle punctuation, to complex ones you can customize. I’ve got one that types OS-X in response to hearing OS ten, and the manual suggests arranging to type your formatted address in response to a phase such as, “address me.”

Recognizing speech is a complicated task, especially since the spelling of many words depends on the context in which they are used. The net result of this is that ViaVoice makes some errors. In the beginning, it makes a lot of errors but you can correct them via a window that allows you to pick from a list of suggestions, say the correct words, or spell them out. As noted above, IBM has given it the capacity to learn from corrections made with this window, as ViaVoice pays attention to when it finds it’s made a mistake and updates your voice model accordingly. Thus, the more you use ViaVoice, the better it gets at recognizing your voice.

Minor gripes and major problems

In many ways, IBM has created a thoroughly Mac-like program. Setup is handled by an interface that resembles the standard Apple Guide, and all windows and palettes conform to Apple Interface Guidelines. Still, IBM did not get some obvious things. ViaVoice demands that virtual memory be on and set to 10 MB above the amount of available RAM even when large amounts of RAM are present. It also greatly increases the use of RAM by the system (by up to 20 MB!), even when extra RAM has been allocated to SpeakPad itself. IBM has limited its transfer capabilities to only five programs when any program that is AppleScriptable can accept text generated in SpeakPad. Some enterprising users have already found how to make Eudora work with ViaVoice, and there’s no reason IBM couldn’t have included a mechanism for doing the same. IBM also claims to support Powermac G3s, but until recently neglected to note that only blue ones are officially supported. This is especially odd, as early iMacs, with nearly identical performance stats to the beige G3s (processor, system bus, etc.), are supported.

The program also seems to be somewhat limited in its capacity to improve its recognition. The ability to analyze documents is limited to a search for unknown words. There is no mechanism for having ViaVoice listen to you reading an entire document, allowing it to recognize either your patterns of speech or the pronunciation and context of “new” words. Even though ViaVoice made some of the same mistakes nearly every time, there’s no way to force the program to listen to you saying the words it gets wrong, and thus no way to force it to improve. Some other annoyances: when dictating, ViaVoice properly gets rid of spaces between punctuation marks and words. Unfortunately, when making corrections, it will put those spaces right back. Finally, it’s worth noting that ViaVoice has crashed several times in a manner that has taken down the entire operating system.

In the end, I had a lot of difficulties using ViaVoice; some of my own making, but in other cases the software forced me to conform to how it wants me to work. The correction window, the main mechanism for improving ViaVoice’s recognition capabilities, is set up to work from the beginning of the document down. Those of us who prefer to catch the mistakes as we make them are out of luck. My reading voice differs greatly from my conversational voice, but the software learns from your reading then attempts to decipher phrases you say conversationally. The slow appearance of text after I started speaking often led me to pause in the middle of a sentence, depriving the software of the context it needed to pick the right words. Finally, whenever anything (man or machine) is mis-hearing what I’m saying, I tend to slow down, speak loudly, and over-enunciate. And that’s so far away from what my reading voice sounds like, ViaVoice never has a chance…

I have mixed feelings about ViaVoice. It does incredible things for program that can be bought for $80. In addition, I have to recognize that it’s a 1.0 release, possibly rushed out in the face of the impending release of a competing product from Dragon Systems. Still, the quirks and annoyances described above frequently leave me frustrated and anxious for an improved release. And I’m beginning to wonder how much time it will take before this program accurately recognizes most things I say. In the end, I’d only recommend ViaVoice to those people with very fast machines who are willing to spend the time to get it to work well with their voices. The rest of us should probably wait for a future release, in the hope that it will be more flexible and have improved functionality.

But despite all, I’m still dictating.

ViaVoice Millenium Edition for the Mac from IBM costs about $80. It requires a PowerBook G3, PowerMac G3 (blue) or G4, or an iMac.

Pros:
Inexpensive, straightforward voice recognition that saves your arms the strain of typing. Has the potential to improve its recognition.

Cons:
Works with only a limited text processor. Requires the latest hardware to run well. Various quirks and bugs, and the potential to improve its recognition seems to be underutilized.

MacMice Rating: 3.5