Breakthrough in Continuous Chinese Speech Recognition Announced by Motorola, Lexicus Division
PALO ALTO, Calif. — November 13, 1996 — The Lexicus Division of Motorola (NYSE:MOT) today announced a major breakthrough in the field of speech recognition research — they have developed a new technology which recognizes continuous Chinese speech.
Running on an industry standard PC, the algorithm can recognize over 10,000 spoken Chinese words and an almost unlimited number of phrases and sentences. In the future this technology should enable Chinese speakers to dictate letters, faxes and e-mail on their PCs.
Motorola's continuous Chinese speech recognition technology will be demonstrated at COMDEX/Fall '96 in Las Vegas Nov. 18-22. Please call Robin Jones at 415/833-8074 to arrange a demonstration of the technology.
Continuous English speech recognition has been the Holy Grail of many American research organizations. "Chinese speech recognition is more difficult than English speech recognition," said Ronjon Nag, general manager and co-founder of the Lexicus Division of Motorola. "Our breakthrough in continuous Chinese speech recognition demonstrates Motorola leadership in developing Chinese-centric technologies."
"We have been pleased at how fast and accurate the continuous Chinese algorithm works," said Elton Sherwin, Lexicus vice president. "We are delighted to get continuous Chinese running on a PC and we are now mounting a major research effort to determine if the Chinese algorithm will work for English and Japanese."
BACKGROUND One out of every five people in the world speaks Chinese. There are 29 languages and dialects used in Greater China, of which Mandarin is the most widely spoken. Mandarin is used in Taiwan, Beijing, Singapore, Central and Northern China. Cantonese, the second most commonly used Chinese language, is spoken in Hong Kong and Southern China.
While there are over 29 dialects of spoken Chinese, there are two standardized systems of printed Chinese, simplified and traditional, with over 20,000 characters. There are approximately 7000 traditional and 5000 simplified Chinese characters in common use. These characters share 408 pronunciations; thus, many Chinese characters sound alike.
TECHNICAL CHALLENGE Since there are few pronunciations for many characters Mandarin uses tones, four tones and one neutral pronunciation, to add more 'sounds' to its spoken language. The tones vary slightly in pronunciation requiring the listener to hear the context of spoken words to understand the speaker's meaning. For example, the Chinese translation of the question "Does the mother scold the horse?" is "Ma1 ma0 ma4 ma3 ma0" spoken using three different tones. Accurately perceiving tones is beyond the capability of many speech recognition algorithms.
Chinese speech recognition algorithms must also distinguish between Chinese homophones. In English it is unusual to find three words which are homophones, e.g., two, too and to. In Chinese, homophones are the norm, not the exception.
CRITICAL POINTS OF COMPETITIVE DIFFERENCE
Several large vocabulary speech recognition technologies have emerged in the last few years. These systems have been discrete systems requiring the user to pause between words. This is difficult in Chinese because written Chinese has no spaces and word boundaries are ambiguous. Motorola's algorithm appears to be unique in its ability to recognize continuous, large vocabulary Chinese speech in real-time on an industry standard PC.
MOTOROLA, LEXICUS DIVISION Motorola, Lexicus Division is one of the world's leading providers of handwriting and speech recognition software for desktop, mobile and embedded systems. Its products include cursive and print recognizers for English and Chinese, and noise-robust DSP-based speech recognition subsystems.
Motorola is one of the world's leading providers of wireless communications, semiconductors, advanced electronic equipment, systems, components and services for worldwide markets. Products include two-way radios, pagers, personal communications systems, cellular telephone and systems, semiconductors, defense and aerospace electronics, automotive and industrial electronics, computers, data communications and information processing and handling equipment. Sales in 1995 were $ 27 billion.
Motorola and the Motorola logo are registered trademarks of Motorola. All other company and product names are trademarks of their respective companies.
CONTACT:
Motorola
Brenda Scariot, 415/833-8062
[email protected]
or
For demonstration appointments contact:
Robin Jones, 415/833-8074
[email protected]