Voice Command software that actually works. Realize Software’s InteractiVoice for Win95/NT4 is the first voice command software for Win95/NT we’ve seen that uses Microsoft’s Whisper speech recognition engine. It’s very, very good (although it is not a voice dictation system like IBM’s US$99 ViaVoice). You don’t need to train InteractiVoice, although those with accents or speech impediments can train the software to improve accuracy. It can open, close, and switch windows, click OK or Cancel, open your Address book or email — even learn to open and close apps or documents you specify. Best of all it’s freely downloadable.
It’s US$39 if you decide you like the fully functional 10-day trial version.
The Computer Pages spoke to Mike Somlo, President of Realize Software, and asked him why the product doesn’t work properly with menu commands in Office 97 or Internet Explorer 4.0. Somlo says, “Floating menu bars present a real challenge since they are not real menus. The only way to access the info in them is through a new Windows API layer that is very hard to use and **extremely** slow. We’re at the point where we are doing heavy-duty Windows tricks to speed this up (by injecting code into the different application’s process space, for example), but the performance may not be acceptable on slow machines.
More later on this…”
Somlo says the company licensed the Whisper speech recognition engine from Microsoft.
Speech technology is also expected to be provided in a future version of Windows CE (and possibly other Microsoft operating systems), although it is not clear whether it will be based on Whisper technology. In May 1997, Microsoft licensed voice recognition technology from Lernout & Hauspie, a Belgian company specializing in speech recognition. L&H also owns Kurzweil, one of the pioneers of voice recognition. Somlo says L&H is not involved with InteractiVoice in any form whatsoever.
According to Somlo, Microsoft has a bunch of developers that have been working on speech recognition for several years together. Most of them had been at Carnegie Mellon, where they had done some leading research on speech recognition. Microsoft hired them and brought them out to Redmond to work on their stuff. Although MS licensed L&H’s technology, I think that was relatively insignificant and probably had a minor impact on this version of Whisper. Microsoft’s more recent venture, investing in L&H (8%) and the Flanders group in Belgium is a much more significant event, Somlo says, and shows that Microsoft is sticking with L&H and their technology and is counting on them coming through to put in their OS.
For those who want their computer to handle dictation tasks, several new titles are worth checking out:
IBM’s ViaVoice (C$159)
Dragon NaturallySpeaking Personal Edition (US$359)
Lotus SmartSuite 97 (the latest update uses ViaVoice technology in WordPro 97)
Of the three, NaturallySpeaking gave us by far the best results, both in terms of accuracy and speed. It also offered better correction options for those inevitable errors that produce such nonsequiturs as the word “lettuce” instead of “let us.”
To correct a word in ViaVoice, we simply said “Correct lettuce.” A list of alternate spellings showed up, which listed ‘let us’ as number two. We said “select two” followed by “click OK” and the change was made. You can say “click File…Save” to save files, as you can command the system to click virtually any button or menu item in almost any program. It worked remarkably well, even when speaking in a normal voice at regular conversational speeds. No longer do you have to talk like this
IBM’s ViaVoice, which also offers continuous speech, had some attractive features. An interesting option in ViaVoice is the ability for the computer to read back existing text files in your choice of voice, complete with an on-screen “agent” of your choice. The package includes male and female cartoon heads, a talking pencil, a robot, and a very Monty Pythonesque talking baby’s head. Our favorite was an innocuous circuit board with a pulsing round light in the center, strongly reminiscent of HAL from 2001: A Space Odyssey.
We found ViaVoice’s feature that allowed dictation inside Microsoft Word 97 greatly slowed down that program and seemingly decreased its reliability. We found the “WordPad” editors supplied by both IBM and Dragon to be better choices for dictating.
The Lotus WordPro 97 update essentially offers ViaVoice dictation features inside that program.
Both NaturallySpeaking and ViaVoice specify a P166 or better and 32MB of RAM as the minimum requirements.
