|
|
|
|
|
       |
 |
|
|
iMessaging Improves Text-to-Speech in Voice Response System by Alex Woodie You might have heard the clunky, unnatural voices when dialing an 800 number to get information about your checking account, or an airline flight. It's no big secret in the industry that the first-generation of text-to-speech technology for integrated voice response systems (IVRs) left something to be desired. But the text-to-speech capabilities of the next release of iMessaging Systems' IVR system, due out next week, holds promise for vast improvements over previous versions.
iMessaging Systems' flagship IVR system, called iVoice, uses a "host controlled" programming technique that allows an AS/400 or iSeries application to command many aspects of the IVR environment. Almost any RPG or COBOL application can be configured to use iVoice as a data source and as a way to present information using telephony. While the brains of iVoice reside in an OS/400 host, its heart resides in an Intel Pentium-powered PC, which runs the Intel cards for rendering audio signals and connecting phone lines and microphones, and the software engines to power text-to-speech conversion and speech recognition. iMessaging Systems provides a properly configured PC as part of the iVoice package. When iMessaging Systems first announced iVoice 2.0 last fall, the PC appliance featured a text-to-speech engine developed by OEM partner SpeechWorks called Eloquent. With the upcoming June 17 announcement of iVoice 2.01, iMessaging Systems is delivering a new text-to-speech engine from SpeechWorks called Speechify. The major difference between Eloquent and Speechify is how they generate words on the fly, says Vernon Harvey, iMessaging Systems' new vice president of development. Whereas Eloquent used computer algorithms to generate the various phonics of specific words, Speechify uses actual recordings of a human voice, and spends its CPU cycles putting those digitized recordings together in the right order. The results, Harvey says, are dramatic. "Through new higher-power computers, we're able to concatenate audio segments together to create a far more natural sounding voice," Harvey says. "It actually has emotion. The older ones were a little flat, monotonous." The new Speechify technology allows iVoice to convert any textual information from the AS/400 or iSeries server into realistic computer generated speech, which should prove useful where prerecorded speech segments are impractical, such as with names, addresses, part descriptions, or other dynamic comments, the company says. The new technology does require a slight boost in PC processing power and memory, and iMessaging Systems now offers Pentium 4 PCs equipped with 256 to 512 MB of memory, instead of Pentium 3s. The improvements in iVoice 2.01 do not in any way affect the processing workload on the OS/400 system, the company says. Speechify is based on research originally conducted by AT&T Labs-Research. The technology has been available for less than a year, and iMessaging Systems is one of the first SpeechWorks OEMs to include it in an IVR system, says Rich Ollari, iMessaging Systems' vice president of operations and the iVoice product manager. "It is truly a breakthrough," Ollari says of the Speechify technology. "We got on the bandwagon very quickly. We're on the leading edge of it." SpeechWorks is so confident that people won't be able to tell a Speechify voice from a real voice that it has created a little online game on its Speechify product page. It's not impossible to pick out the real voice from the computer-generated voice, especially if you concentrate on the inflection points. But the Speechify technology seems to do a remarkable job of mimicking the speech patterns of real people, so the game poses a bit of a challenge. This reporter scored seven out of 12 right, slightly better than average (and chance, for that matter). iVoice 2.01 also features improvements to its speech recognition capabilities, which iMessaging Systems also gets from SpeechWorks. With this release, iVoice uses a new SpeechWorks' technique called "dialog modules" that is supposed to improve the accuracy of speech recognition. According to iMessaging Systems, the new speech recognition capabilities will allow iVoice to recognize "yes" or "no" responses, spoken digits, such as telephone numbers or credit card entries, as well as item list keywords, such as "arrivals," "departures," or "flight schedules." Other improvements delivered with this release include a new call log reporting module for OS/400 for tracking calls and iVoice usage; a dynamic redundant takeover feature that replaces the primary iVoice controller (the PC side) when it goes down; and a new "configurable option" that allows the upload, storage, and synchronization of iVoice voice files to the OS/400 system, removing the need for voice file backup and improving the speed with which a new or replacement iVoice server can be brought on-line. iMessaging Systems was founded in late 2000 in New Hampshire by Karen Sedlar, who was the head of IVR vendor RMTi. iMessaging Systems experienced a downturn in business last year following the September 11 terrorist attacks, and it was slow to recover this year. But business is starting to look up, officials say, with a customer recently signing the largest contract in the company's short history. Pricing for iVoice starts at $16,500 for a two-line unit and scales all the way up into the T-1 line territory. For more information, visit www.imessagingsystems.com.
|
Editor
Contact the Editors |
|
Last Updated: 6/11/02 Copyright © 1996-2008 Guild Companies, Inc. All Rights Reserved. |
ASNA