ScanSoft updates voice software

Software maker ScanSoft is set to announce on Tuesday a new version of its OpenSpeech Recognizer software for running voice-activated phone systems.

The OSR software typically is used for customer service lines and corporate phone systems to direct calls and help callers retrieve information.

ScanSoft is one of the leading makers of voice recognition software, a growing field, attracting interest from computing giants such as IBM and Microsoft.

The new version 3 of OSR is the company's first product to incorporate technology from SpeechWorks, a voice software specialist ScanSoft acquired about a year ago.

Enhancements in the new version include improved natural-language capabilities that free customers from having to bark out orders in two-word fragments. "It lets you speak in full sentences, which is a lot more natural," said Steve Chambers, president of the company's SpeechWorks division.

The new release also improves name recognition and boosts language support to cover 40 tongues. The software also has learning capabilities, so it gets better at recognizing and interpreting an accent, the more it encounters it. "That's a very important capability for a multicultural society like the U.S.," Chambers said.

OSR 3 also includes the usual boosts in accuracy, with particular improvements for calls originating from cell phones, where dropped packets can often stymie voice software on the other end of the call. "It used to be (that) you could count on everyone using wireline for calls, but now you have to adapt for different types of connections," Chambers said.

OSR software development kits are available in Windows, Linux and Solaris versions, and sell for $4,995.