Intel develops Chinese voice-recognition platform for future applications
Software will run on x86- and StrongARM-based processor products
By Mark LaPedus
Semiconductor Business News
(04/13/01 17:41 p.m. GMT)
BEIJING -- In an effort to optimize its microprocessor lines for next-generation applications, Intel Corp.'s subsidiary in China is developing an advanced voice-recognition software platform that promises to solve some major problems in the industry.
Design to run on its x86- and StrongARM-based microprocessor lines, Intel's distributed speech-recognition (DSR) software technology supports both English and Chinese languages--particularly the Mandarin dialect.
The DSR technology can also be applied to support other foreign languages as well, according to Peter Liou, director of the Intel China Research Center. Based in Beijing, the research center is part of Intel's China subsidiary, Intel China Ltd.
Eventually, Intel hopes to develop more advanced speech-recognition technologies, like universal translation machines, Liou said. These applications will require processors that run at speeds of 10 GHz or faster, he said.
"Speech-recognition is a CPU-intensive application," Liou told SBN in a recent interview at the company's Beijing research center. "What we're doing is developing technology for tomorrow's microprocessor applications."
But Intel has an usual business model for its voice-recognition software. Fearing that it will compete with its microprocessor customers, Intel does not plan to sell the DSR software platform on the open market, Liou said.
Instead, the company will license--or give away--the technology to third-party developers, such as IBM, Lernout & Hauspie, and others, Liou said.
IBM, Lernout & Hauspie, and others separately offer voice-recognition software products for use in Chinese-language applications. These software products are typically optimized to run on a range of CISC- and RISC-based processors, such as Intel's x86, Sun Microsystems' Sparc, and PowerPC architectures.
For these developers, Intel licenses or gives away what it calls "low-level software functions." As a result, the software developer is able to optimize a product on a particular platform, notably the x86 or StrongARM, he said.
Intel's work in speech-recognition is important for other reasons. Speech-recognition technology promises to solve some major problems for Chinese-language populations in China, Hong Kong, Taiwan, and other regions.
At present, the Chinese language consists of over 10,000 characters. As a result, the Chinese-language keyboard on a PC is complex and cumbersome for users. Even the fastest typists in China can only type some 40 characters a minute on a Chinese keyboard, analysts said.
Speech-recognition could eradicate the need for the Chinese keyboard on a PC, but there is still some major problems with the technology. "Speech-recognition is in its infancy," Liou said. "The problem with speech-recognition is that it's only 90-to-95% accurate."
While the jury is still out for speech-recognition products for the PC, the technology is taking off in non-PC applications, like voice-portals, call-centers, and others. "Speech-recognition is taking off in applications where you don't need 100% accuracy," he said.
Intel hopes to solve the accuracy--and the noise-related problems--associated with speech-recognition. Targeted for call-centers, portals, and other applications, Intel's DSR software is designed to reside on both the server and a piece of end-user equipment like a cellular-phone or PC.
By using a proprietary speech-engine technology, voice patterns can be digitized and distributed through the network. "The technology builds a model of your voice," Liou said.
Intel's technology is compliant with an industry specification for voice-recognition. Dubbed Aurora, the specification is still in the state of flux, however. "There is a standard, but there is really no implementation for the standard," he added.
