|
|
![]() |

LOG ON--AND LEARN A LANGUAGESpeech recognition is improving, so these programs can perfect your parlez-vousThe computer asks for the German word for ``excuse.'' I say ``Die Entschuldigung'' into the microphone. ``Korrekt,'' responds a brisk Teutonic voice. This is not the language lab of my youth. Instead, it's Learn to Speak German, an imaginative, if flawed, use of speech-recognition technology. So far, most efforts to commercialize speech recognition for personal computers have focused on letting you dictate a letter or memo to a computer, which converts your speech into text. This is an obvious use of today's technology and should cheer the hearts of the keyboard-phobic among us. But the real payoff from speech recognition may come with more creative applications, from telephone services to education. GOOD GRADES. Language training is a natural for speech software. The $100 Learn to Speak German program is from Learning Co. (800 852-2255). Like its sister programs, Learn to Speak Spanish, Hablemos Ingles, and Learn to Speak English, it offers drills essential to language learning. While earlier multimedia language programs let you hear a foreign language, these offer instant feedback on your pronunciation as well. Not only did the program let me compare my German to that of a native in vocabulary and pronunciation drills, but a meter graded my performance from ``tourist'' to ``native speaker.'' The CD-ROM program, plus a 400-page-plus workbook, are designed to cover at least a year of college-level study. The speech recognition, built around software from Belgium's Lernout & Hauspie Speech Products, is impressive but not perfect. In the more advanced drills, Learn to Speak German was smart enough to recognize that there's more than one correct way to word an answer, and it accepted several variations. Yet in vocabulary drills, the program often had trouble understanding short words, such as prepositions, no matter how carefully I spoke. Unfortunately, the program suffers from the bugginess that infects many Windows multimedia programs. I set it up on four different computers with Win 3.1 and Win95. Only one installation succeeded on the first attempt. The program crashes occasionally without warning, often producing such befuddling error messages as ``Script Error. Handler not defined.'' A Macintosh version (included on the same CD) ran flawlessly but lacks the speech-recognition feature that makes the program so attractive. SOFTWARE FLOOD. Learn to Speak wants a fast computer. Although the specs say it will run on a 33-megahertz 386DX, I found that a 90- Mhz Pentium is the slowest machine that avoids unpleasant delays while the computer analyzes your speech. A decent microphone also can greatly improve recognition accuracy. I had good results with a $30 battery-powered Koss M/21. Language programs and dictation software from companies such as Kurzweil AI, Dragon Systems, and IBM are the leading edge of what is likely to turn into a flood of speech-enabled software. Microsoft Corp. is designing a standard way to add speech features to Windows 95 and Windows NT programs. This being the software industry, Microsoft's Speech Application Programming Interface (SAPI) effort is being challenged by the Speech Recognition Application Programming Interface Committee (SRAPI). The SRAPI group, headed by Novell Inc. and including the leading speech-recognition-technology companies, is developing its own standard for a variety of computer systems. One hopes that some benefit will come of this alphabet soup. In the meantime, the Learn to Speak programs, bugs and all, are an interesting preview of things to come and a good way to learn a new language or brush up rusty skills. BY STEPHEN H. WILDSTROM
|

Updated June 14, 1997 by bwwebmaster
Copyright 1996, by The McGraw-Hill Companies Inc. All rights reserved.
Terms of Use