CNET también está disponible en español.

Ir a español

Don't show this again

Culture

Microsoft, IBM tout speech software

Speech recognition, one of the computing world's long-deferred dreams, will be highlighted in separate initiatives from the two companies next week.

Speech recognition, one of the computing world's long-deferred dreams, will be highlighted in separate initiatives from Microsoft and IBM next week.

On Monday, Microsoft, along with representatives from Cisco Systems, Intel, Philips and SpeechWorks, will hold a press conference at its Mountain View, Calif., offices. The companies will announce a collaborative effort to lay the groundwork for accessing the Web through voice commands, according to a statement from Microsoft.

The idea behind the effort is to come up with practical ways to "provide consumers with multiple means of accessing information anytime, anywhere over a number of different devices--including PCs, wireless (handhelds), mobile phones and telephones--over diverse networks," according to Microsoft.

Speakers at the event include Alistair Woodman, director of marketing in the voice technology center at Cisco; Howard Bubb, vice president of the telecommunications and embedded group at Intel; and Kai-Fu Lee, vice president of the natural interactive services group at Microsoft.

A Microsoft representative would not release further details about the announcement.

IBM, meanwhile, will unfurl the technical details behind the PowerPC 405LP, a chip for handhelds that contains dedicated circuitry for performing speech recognition.

"That is kind of a breakthrough for handhelds. When you try to do speech recognition on a Palm or a Pocket PC, you are turning everything else off," said Richard Doherty, director of research for The Envisioneering Group, a Seaford, N.Y.-based research firm. "Speech recognition so far is sound recognition. Until you can tell it is a sentence, it is very clever sound recognition."

IBM already produces speech recognition software for servers, handhelds and PCs.

Speech recognition has been a staple of technology news for years, but often for the wrong reasons. Computer visionaries have often discussed a future where individuals access databases by speaking to machines that can understand the syntax and linguistic context behind the spoken word. Microsoft Chairman Bill Gates, among others, has made speech recognition a staple of his trade-show speeches.

But speech-capturing techniques are still prone to errors. Speech recognition programs also require fast, powerful processors.

As a business, speech recognition has been no joyride either. Lernout & Hauspie, one of the pioneers in the field, has been in a drastic tailspin since last year when it filed for Chapter 11 bankruptcy protection. Since then, government investigators have alleged that the company falsely booked $100 million in orders. Dragon Systems, once a lead competitor, was bought by L&H before the trauma.

The downfall of L&H has made IBM the dominant force in the market.

"The only real competition for IBM is coming out of the universities," Doherty said. "IBM has the largest number of speech scientists on the planet."

Still, over the past two years, speech-to-Web applications have been popping up with greater frequency. Airlines' information lines that can retrieve arrival times through voice commands, instead of the "press-one-for-flight-times" button commands, are examples of such applications. In July, IBM was part of a group of companies that proposed a standard for voice-Web access called VoiceXML.

Microsoft and Intel have also included hooks into the Windows XP operating system and Pentium 4 chip, respectively, for running speech applications.