TO ALL:
I stumbled on the following news while browsing. I have no axe to grind but thought this thread would find this interesting:
Hear This!
Voice recognition will simplify life for Asian users
By Charles Bickers in Hong Kong
December 24, 1998
W hen you're developing technology that requires high-performance PCs and
may be the next "must-have" item for computer users in China, you can count on
at least one big supporter: the company that sells most of the world's
microprocessors.
"Intel loves us," says Louis Woo. "They know this is important."
The 52-year-old American managing director of Singapore-based software
company Lernout & Hauspie Asia is betting big on Chinese-language voice
recognition. He believes it will radically change the way a whole new generation of
Chinese, Koreans and Japanese work, play and above all, communicate.
Sitting in front of his laptop computer, Woo speaks Cantonese with a slightly
formal tone into a headset microphone, dictating a story from a newspaper into the
PC using his company's new software product, SP2. After less than a minute, he
finishes and corrects a few characters by drawing the ones he wants using a
pen-like device. "It would take most Chinese speakers forever to type that," he
says.
Creating Chinese characters on a keyboard can be done in several ways, all of
which are slow and unnatural compared with typing Western languages. In fact, a
reasonably computer-literate Chinese speaker can type only 10-20 characters a
minute. With voice recognition, however, the same user can input 100-150
characters into a word-processor document or e-mail message in the same time.
The potential for speech technology is huge, say many in the computer industry.
In November, an international forum in Beijing drew around 15 companies offering,
or about to offer, speech-based software. A similar number of university research
departments--particularly ones from mainland-Chinese universities--showed off
their own work. As Patrick Gelsinger, Intel's general manager for desktop systems,
explained at the forum: "The problem is that the keyboard that we know today
was not designed for the world's population. There are 1.2 billion Chinese who
were not trained in the Roman characters and who were never taught how to use a
keyboard."
This is why Woo stresses that although speech recognition is regarded as a novelty
elsewhere, for character-based languages it is a must. "In the next 10 years you're
going to see more solutions using speech coming from Asian countries than the
West. The simple fact is that they have a far more compelling reason to use
speech."
Woo has been involved with developing voice-recognition software for about seven
years. But his excitement about the technology is higher than ever: Only now is the
power available in a normal PC proving capable of handling the tough job of
comprehending spoken words.
A computer can't understand the brisk pace of conversational speech unless it's
equipped with complex software--plus, the task is very demanding of PC
hardware. The ability to identify the spoken word requires that a computer's
processor matches the sound against thousands of patterns stored in its memory.
Consistent and clear speech helps the computer, as does superior audio hardware,
but a fast processor is key.
Intel's next generation of Pentium microprocessors, code-named Katmai, will
feature improvements to make speech recognition easier. To be launched in the first
half of 1999, Katmai will be faster, running at a speed of 500 megahertz, compared
with today's 450-megahertz, top-line Pentium II processors.
Intel is embracing speech recognition wholeheartedly because it hopes the
technology will be a new "killer app"--one that consumers feel is a "must-have,"
which will then drive demand for Intel's highest-performance processors.
Woo demonstrates SP2 on a relatively slow laptop computer. While the software
performs well, there is a slight, but noticeable, delay--say, a couple of
seconds--between speech and the on-screen display. "With a top-line computer,
there is hardly any delay," says Woo.
If he's embarrassed, he shows no sign of it. Fixing the few errors is
easy--highlighting the mistakes with the pen brings up a menu of alternative
sound-alike characters. If none of those is correct, writing the character takes little
time.
For now, the product is aimed at the business and government markets. Because of
the limited vocabulary, "I don't recommend it if you're writing a book just yet,"
says Woo. Amassing a business vocabulary of 50,000 phrases, which along with
individual characters are the building blocks of the language, was a big enough
challenge, but Woo is sure that it won't be long before more characters are added. In
addition to Cantonese software, the company also sells Mandarin versions. And
although it doesn't plan to add Japanese or Korean in the near future, they are part
of the company's long-term vision.
Woo, born in Macau and raised in Hong Kong, first came across speech technology
while working as a manager for Apple, maker of Macintosh computers. "When I
first saw this technology, I said, 'This is it, I've got to have it.'" Looking back,
though, he acknowledges that while the technology then was "cutting edge," it fell
short because the computers "just weren't up to it." Accuracy was 70%-75%,
which left users using pen and keyboard frequently. By contrast, L&H Asia says,
SP2 is 90%-95% accurate.
Apple's first version of Chinese-language voice-recognition software, released in
1996, required users to train their computers for four hours before using it in
earnest. Now, a user need only read 25 sentences, which takes about five minutes,
before the program is ready. Another problem that has been resolved was that the
program couldn't easily understand Chinese spoken in a natural way. But software
issues aside, the real speech-stopper was the limitations of the microprocessors
then available.
In the mid-1990s, Woo and 10 others were laid off from Apple's Asia headquarters
in Singapore. Convinced of the utility of speech recognition for Chinese, Woo and
colleagues formed a software company, Asiaworks, and licensed Apple's
Chinese-language technology so they could develop and sell it for the more widely
used Windows-based PCs. In late 1998, Lernout & Hauspie, a U.S.-listed,
Belgium-based speech-software specialist, bought 49.5% of Asiaworks.
Woo's road may already have been long, but he anticipates it stretching still further
ahead. He hopes to see the day when voice recognition is employed in everyday
devices such as telephones and organizers, where typing Chinese characters is
hopelessly arduous. Some products, such as Philips' Genie mobile phones, already
use the technology for commands such as dialling stored numbers. Woo says it's
just the beginning. Anyone who uses programs that require typing a character had
better listen up.
Any comments? This is the main reason I have NOT bought into companies similar to Zi. |
