A foreigner might find it easier to learn about Beijing even if
he doesn't speak Chinese during the 2008 Olympic Games in Beijing.
For example, if he wants to taste Chinese food and yet doesn't
know where he should go, he may use his roaming cell phone to dial
a number and speak his inquiry. Seconds later, he will get ample
information in English from a computerized polyglot.
Pan Jielin, who works on employing speech recognition
technologies in Olympics-related service, told Xinhua Beijing will
be the first city in the world to extensively offer the
multi-lingual computerized information service.
Beijing tourism authorities estimated Beijing will host at least
550,000 foreign and 2.2 million domestic visitors during the
Olympic Games.
"We now only have Chinese and English services, but will expand
to other languages including French, German and Spanish," said Pan,
associate director of Thinkit Speech Laboratory at the Chinese
Academy of Sciences (CAS) Institute of Acoustics.
Pan's lab has developed the embedded multi-lingual speech
recognition engine, which picks up acoustic features of human
speech, coverts sound signals to bytes, compares discourses of
speakers with various syllables in different languages, and
optimizes match-ups from algorithmic processing. In a matter of
seconds, speakers could get response from the system.
"The core technology of speech recognition applies to any
language if we get big enough speech databases," Pan said.
"We are very competitive in processing the Chinese language
because we're able to get excellent Chinese databases, including
those of dialects," said Zhao Qingwei, the lab's chief technology
officer, citing that his lab bought native-speaking English
databases from American companies.
Although quite competent at recognizing mandarin Chinese spoken
by TV newscasters, Pan said the system performs undesirably while
keeping notes of casual conversations.
"We're quite confident of recognizing more than 90 percent of
speeches of certain topics, such as road and traffic information,
Olympic competition results, Olympic venues information, and
weather information," Pan said.
Pan's team is focusing on improving recognition rate of spoken
languages and reducing noise affects.
The lab got contracts from Capinfo Limited Company, the solely
authorized multi-lingual service provider for the Beijing Olympics,
for equipping telecommunication platforms and information booths
along main Beijing streets with the speech-recognition engine.
China Mobile, the country's largest monopoly over mobile
telecommunication, has already employed the recognition technology
in its new valued-added service, Coloring Ring Back Tone.
Subscribers could easily find popular melodies, even if they don't
know their titles, by just humming a few segments.
"We look forward to earning smart returns from the profits we
might share with China Mobile," Pan said.
The speech and melody recognition technology is also welcomed by
entertainment places such as Karaoke arcades. Karaoke fans, in some
places where the recognition system has been installed, are able to
search songs by singing a little of them.
(Xinhua News Agency September 11, 2007)