APEX Voice Communications Integrates
Nuance^® Speech Recognition Software.

New Module Allows for Increased Hands-Free Access
to Enhanced Network and Enterprise Services

SHERMAN OAKS, CA, APRIL 2, 2002 – APEX Voice Communications, a leading supplier of enhanced services and billing solutions to network service providers and enterprises worldwide, today announced that it has enhanced its speech recognition offering by integrating Nuance (Nasdaq: NUAN) speech recognition software into its OmniVox® Intelligent Call/Media Processors. With the newly added Nuance-specific command icons, OmniVox developers worldwide can now easily create a greater variety of voice-driven wireline and wireless (hands-free) applications using the built-in service creation environment. And by incorporating Nuance’s Natural Language and N-best processing features, callers can interact with the system more naturally and with higher accuracy.

Eight new command icons have been added to the service creation environment of OmniVox giving developers the flexibility to integrate Nuance speech recognition software at any point during a call flow. These commands include recognizing a series of spoken digits, recognizing a series of spoken digits and letters, as well as taking advantage of natural language capabilities, which include "understanding" whole sentences and picking out the important bits of information and making them available to the application in a useful way. Additionally, all these commands are "DTMF-enabled", where they are capable of detecting DTMF key inputs in addition to speech.

"For over a decade now, OmniVox has had some type of limited-vocabulary speech recognition capability," stated Elhum Vahdat, Executive Vice President of APEX Voice Communications. "However, it has never had the flexibility to allow the caller to speak naturally and in complete sentences, until now with Nuance."

Natural language speech recognition allows the caller to enjoy faster and more efficient phone-based interactions since the caller speaks in complete sentences rather than in single words. Because every caller has a unique way of speaking, the natural language engine has the ability to select specific words from within the spoken sentence and place them in "slots" after which it performs a specific function based on the combination of these captured words. For example, when a caller says, "Transfer 500 dollars from checking to savings" the natural language engine selects the following words and places them in the appropriate slots and then performs the spoken action -

Another feature that has been integrated into OmniVox is Nuance's N-best processing. Traditionally, speech recognition engines would perform the action that met the single highest level of accuracy based on the recognized word. However, with N-Best, the speech recognition engine can return more than one possible recognition. This is especially useful for DTMF and alphanumeric recognitions where the accuracy rate tends to be lower. For example, a string of numbers that represents a telephone number can be tested against a list of possible area codes and exchanges, with any impossible data being eliminated. Then any recognitions which remain can be confirmed by the caller with a yes/no menu, dramatically increasing the level of accuracy.

Today Nuance supports 26 international languages allowing OmniVox developers to deploy voice-driven solutions worldwide. Currently, Nuance is available on OmniVox for Windows 2000 and will be available for OmniVox for Linux by the end of the second quarter.