ViaVoice Tested And Found Wanting

Voice Power Pro a better choice in Mac dictation software, at least for now, and MacSpeech's iDictate is coming

I really wanted to like IBM's Viavoice. I have problems with neuritis, that affect me all over, but especially in my arms and hands, and that sometimes makes typing pretty uncomfortable.

Having said that, I'm not that crazy about the dictation concept. I operate better in "mind-to-fingers" mode, as it were, and I find thinking out loud a difficult adjustment. However, I had experimented with an older version of Voice Power Pro -- enough to recognize that dictation has possibilities.

So when a particularly bad episode of neuritis hit back in mid-March, I decided to try a copy of IBM's ViaVoice Millennium Edition for the Mac.

Why not get the latest version of Voice Power Pro? Well, ViaVoice is newer, and boasts continuous speech recognition, versus long-in-the-tooth Voice Power Pro, which is a development of Dragon Systems' old PowerSecretary for the Mac -- a discrete-speech program that...requires...you...to...dictate...like...this. I thought I would opt for the newer technology, and at $89.99 ViaVoice is also slightly cheaper than Voice Power Pro. However, I discovered that this is one of those times when you get what you pay for.

Unfortunately, ViaVoice is several bricks short of a load in the user-satisfaction department. It distresses me to have to say this. I am delighted that IBM decided to support the Mac with its speech recognition software, especially since Dragon Systems recently bailed out of the Mac market a second time. However, I found ViaVoice to be both excruciatingly slow, and maddeningly unstable.

Now, my PowerBook G3 233 (512k cache) is minimum Mac hardware supported by ViaVoice, so I wondered if maybe IBM had pushed the support envelope one step too many. I emailed my son, who also recently installed ViaVoice on his 333 MHz Lombard PowerBook. He reported that it is painfully slow on the Lombard as well. I mentioned this in my Road Warrior column on MacOS Daily last week, and received this letter from reader Bruce Belton:

"Charles,

"Read your article with interest and thought I'd share with you that ViaVoice is no better on a 500mhz Pismo than it was on my 233 WallStreet.

"I hoped for some improvement but for me (a good typist also) it is just not a usable product even with the zippiest PowerBook.

"G4 maybe, hmmmmm."

Maybe, but my son says that ViaVoice is sluggish even on a G4 system. Altivec ("Velocity Engine") optimization might help, but I don't know if that would be possible in what is essentially a port of a PC product.

IBM has posted a version 1.0.2 upgrade patch for ViaVoice, which I downloaded and installed before using the program. The patch fixes and improves a number of issues with ViaVoice version 1.0, including:
• ViaVoice 1.0.2 is more tolerant of background noise on certain systems
• It is now possible to train words with dicriticals/accents and special characters
• For newly created users it is now a default to single space after period
• Analysis of text documents is now faster (the mind boggles how slow it must have been in version 1.0)
• Large SpeakPad documents are handled better when memory is depleted
• Extra spaces between corrected words and punctuation marks are now eliminated
• Miscellaneous 'Scratch That' and 'Undo' problems have been corrected
• There were some instances where SpeakPad could not be opened; this has been corrected. (I hope so!)
• Miscellaneous other fixes

There are some good things about ViaVoice. It's easy to install, and the setup interface is clear and intuitive. The print manual is considerably less so, but at least there IS a print manual, albeit a thin one.

Before you can use ViaVoice for dictation, you must train it to recognize and analyze your voice by reading it some short stories that are provided. I read just the two selections from Robert Louis Stevenson's "Treasure Island," although some suggest that for optimum accuracy, one should read the entire list of selections. After you finish reading, the program needs half an hour to an hour of unattended computing time to analyze the patterns of your voice.

Accuracy was not my beef with ViaVoice. In fact, I found it quite amazingly accurate and intuitive even after just the initial "Treasure Island" training.

You can also have ViaVoice scan the documents already on your computer and import words discovered that ViaVoice doesn't already have in its dictionary.

The biggest functional difference between the way ViaVoice works as compared with Voice Power Pro, is that while the latter allows you to dictate directly into windows and text fields of a wide variety of Mac applications, with ViaVoice you must dictate into the program's own, dedicated, text editor -- "SpeakPad," and then transfer your work to the desired destination application either via the clipboard, or directly into one of five supported programs -- America Online; Outlook Express; Netscape; AppleWorks; and Microsoft Word. The only one of these I use regularly is Netscape, and it only for browsing, not email, so the clipboard was effectively the only mode for me.

And one wants to make sure to transfer early and often. Perhaps the most disheartening experience I had with ViaVoice was spending two hours dictating, saving the work to a ViaVoice document every second paragraph or so. Just as I was finishing up, literally on the last sentence, SpeakPad crashed, rendering all my work unrecoverable, including everything I had saved before the crash. I had given SpeakPad three times the minimum specified memory and half again the recommended partition, so that should not have been the problem. I have never before had a Mac application crash and render previously saved document contents irretrievable.

SpeakPad is also agonizingly slow at opening its document window and in doing saves. Even more annoying is that its "continuous speech recognition" is in real world use (at least on my G3 233) slower than Voice Power Pro's pause-between-words motif, and you have the impression that the processor is constantly struggling to keep up with the program's demands.

If the program misinterprets a word, there is a correction window that provides several correction options. This works well enough, but you guessed it -- in slow motion.

I found that the Andrea NC-71 microphone that ships with ViaVoice works acceptably well, although others have reported problems associated with the power adapter port on the G3 PowerBooks. The head set is comfortable, and the microphone seems to work well and accurately enough. Included are small colored snap-on shells in the five iMac colors. I did not use these because I don't have an iMac, however, I have heard that your choice of color is irreversible, at least without breaking a shell to get it off. ViaVoice will reportedly work with several other brands of microphones.

ViaVoice is a lot better than no dictation software if you can't type or if you have repetitive stress injury. However I can type roughly three times faster than ViaVoice will copy my dictation, and I am not a particularly fast typist. For now, at least until MacSpeech ships it's iDictate program, expected later this year, I would recommend Voice Power Pro as the better choice even though it is slightly more expensive and older technology. While researching this article I tried the Mac OS 8.1 compatible version of Voice Power Pro again, and it is definitely faster in actual use than ViaVoice. The developers say that version 3 allows even faster dictation, and is now one of the fastest dictation systems available for the Macintosh. They are being modest, since the only other dictation system currently available is ViaVoice, and it is definitely slower.

ViaVoice seems to have the edge in voice recognition accuracy, at least during the short period that I used Voice Power Pro, although the difference is not dramatic. However I was gratified by how much Voice Power Pro's recognition improved as I used the program even during a short session.

Another featureVoice Power Pro 3 offers that ViaVoice does not, is reasonably comprehensive "keyboard emulation," -- that is: type anything you say into virtually any text based Mac application. You can also control AppleScriptable software by speech. As noted, ViaVoice requires you to dictate into SpeakPad and then transfer your dictated text via the clipboard or to one of 5 supported programs, which is passably OK for writing, but does not work so well with spreadsheet or database programs. You can also use Voice Power Pro to control most Mac functions from the Finder.

Unlike ViaVoice, with its very heavy, G3-minimum, processor demands, Voice Power Pro 3 will work with most PowerMacs and Mac OS clones from the 6100 on. I found that it worked tolerably well even on a 100 MHz PowerBook 5300 -- indeed faster on that old machine than ViaVoice is on my 233 MHz PowerBook G3. Once you have VoicePower Pro trained to recognize your voice patterns, it reportedly can achieve accurate dictation speeds of 50-100 words a minute. The initial training process is more cumbersome and boring than with ViaVoice's, and takes about one hour, but the program continues to learn as you use it, and speed and accuracy both improve over time. A dictionary of 120,000 words is included, to which you can add your own custom dictionaries.

The program comes complete with a high quality Vxi microphone with an attached pre-amplifier. The software has been optimized for use with this Microphone. The Voice Power Pro 3 US version is Mac OS 7.5.5 - Mac OS 9 compatible.

Voice Power Pro 3 is available from MacMall, and is distributed by Ingram Micro (SKU number 372 310). It is now selling at MacMall for $109.99, a considerable reduction from its former price. (Three years ago, its direct ancestor, PowerSecretary, was selling for $2,500.)

For more information, visit:
http://www.voicepowerpro.com/

In summary, at least at this stage of ViaVoice's development, I have to give Voice Power Pro the edge in both value and performance by a substantial margin. (Please note that I am assuming here that Voice Power Pro 3, which I have not yet tested, is at least as good or better than Voice Power Pro 8.1, which I have tested). However, I do hope that IBM will continue developing ViaVoice for the Mac. The accuracy even after just the initial training is amazing. What needs work is speed, stability, and the ability to dictate into any text field. For now, Voice Power Pro is faster, more stable, and more versatile.

iDictate, currently under development by MacSpeech Inc., will combine continuous speech recognition (using the FreeSpeech 2000 speech engine licensed from Philips Speech Processing) with the ability to dictate text anywhere and Mac Finder commands in a speech recognition program coded from the ground up for the Mac only. If it can successfully combine the virtues of ViaVoice and Voice Power Pro, and avoid serious bugs, MacSpeech should have the premier dictation application for the Macintosh platform.

"Our products will behave the way you would expect a Macintosh dictation product to behave," says Chuck Rogers, of MacSpeech, Inc. "You can talk anywhere you can type. Another 'MacSpeech' advantage is our extensive command and control and AppleScript support."

***

IBM's ViaVoice Millennium Edition for the Mac supports Mac OS 8.5.1, 8.6 and 9.0. The software requires a 233 MHz Power PC processor or faster, 48 MB RAM, 200 MB of hard drive space and a CD-ROM drive. This software runs on iMac, PowerMac G3 and G4, and PowerBook G3 computers, and requires an audio input jack that is compatible with Andrea NC-71 microphones. The suggested retail price for the software is US $89.95. For more information, visit: http://www.ibm.com/software/speech

Charles W. Moore

.

Friday, 25-Jul-2008 07:00:24 EDT