New Programs Take Better Dictation

Jon Pepper

Though not quite there yet, the latest speech-recognition programs are closer than ever to reaching the ultimate goal of speaker-independent, continuous dictation. The newest programs allow you to speak naturally when giving your PC commands such as "Open file," but they still require you to pause slightly between words during dictation. However, the length of that pause continues to decrease, and, as mentioned in "I'll Talk To You Soon" (March BYTE), vendors expect to reach that continuous-dictation holy grail in the next few years. Until then, the ever-increasing power of PC hardware, combined with more-sophisticated voice-recognition software, makes voice recognition a viable choice for many users.

"Most of these products are in their second or third generation ," says William Meisel, head of TMA Associates (Encino, CA), a market-research firm covering speech technolog y. "The market is now mature, and many of the previous impediments have been removed."

For example, one major roadblock, the requirement of a dedicated card, has been eliminated now that these programs work with most major sound cards. Also, the major programs no longer require lengthy training sessions, although accuracy usually improves with an optional training session and continued use.

The three key players in the PC-based voice-recognition market -- Dragon Systems, Kurzweil Applied Intelligence, and IBM -- all have roughly equivalent capabilities, but with inevitable differences. For instance, IBM's new VoiceType 3.0 program, a subset of which will appear in the next version of OS/2, enables you to dictate a program and defer its correction to a later date -- or even to someone else, such as an assistant. And the latest version of DragonDictate can re ad back any file or portion of a file that you designate.

Nevertheless, barriers prevent these current products from reaching mass acceptance. For example, hardware requirements are relatively hefty. Kurzweil and Dragon say that their products will run on a 486 but recommend a Pentium. And IBM requires at least a 90-MHz Pentium-based PC with a 256-KB L2 cache and 16 MB of RAM. Also, audio files created by IBM's delayed correction can quickly consume hard disk space.

These requirements, plus the elusiveness of continuous dictation, limit much of the potential market to specialized users, such as people who lack typing skills or who suffer from a repetitive-stress problem. Says Meisel: "People just aren't strongly motivated to use a microphone yet if they're comfortable with a PC and a keyboard."

---

Win95 Dictation Highlights

Product/Company
                                       
Price


-----------------


Highlights


===================================



DragonDictate 2.5 for Windows
               Personal Edition (10,000-word
Dragon Systems                              active vocabulary), $395; Classic
Phone:    (800) 825-5897                    Edition (30,000 words), $695; 
Internet: 
http://www.dragonsys.com
          Power Edition (60,000 words), $1695

New text-to-speech feature reads text back to you; voice macros for
Netscape Navigator; 10,000- to 60,000-word, fully loaded, active
vocabulary; continuous speech for commands; pop-up window aids
on-the-fly correction; NT support; no training required.


IBM VoiceType 3.0 for Windows 95
            $699
IBM
Phone:    (800) 825-5263
Internet: 
http://www.software.ibm.com/workgroup/voicetyp


Delay
ed and delegated correction let you edit a document later or
send a copy of a file with your recorded voice to assist someone else
in correcting; no training required; continuous speech for commands.


Kurzweil Voice Release 2.0
                  $695
Kurzweil Applied Intelligence
Phone:    (800) 380-1234
Internet: 
http://www.kurz-ai.com


No voice training required; 30,000- to 60,000-word vocabulary;
supports direct dictation within over 30 Windows applications.