Page Two

By Michael Caton  |  Posted 2004-08-30 Print this article Print

Webster said the biggest challenge in making the transition to voice hasnt been writing code but grasping the technology.

"Our biggest challenge ... has been understanding the technology, understanding what is a prompt, what is a grammar and how does speech recognition work," he said.

Two Woodforest application designers spent several months learning about voice technology and took classes in voice user-interface design offered by Enterprise Integration Group Inc., a San Ramon, Calif., company specializing in IVR applications.

The banks development team, comprising one full-time and one part-time person, needed only two weeks to write the first draft of the application. The team used the bank staff to stress-test the system, which is slated to be deployed soon.

The application was designed as a directed, rather than an open, speech system. As a directed system, the application prompts the user with a question and expects a short answer in return. For example, when the application asks a customer, "Would you like your balance or transaction information," it expects a response such as "balance" or "transaction."

This approach allowed the Woodforest developers to build a relatively small grammar for the application. Webster said the company plans to expand the application over time to encompass a broader grammar, which will allow for presenting users with open-ended questions such as, "What would you like to do?"

Webster said a significant benefit to moving from the DTMF system to Microsofts Speech Server is improved reporting. With the DTMF system, only a small amount of call data can be tracked. Speech Server, in contrast, collects a large amount of information about a wide variety of data points, including call duration, features being used and recognition-related problems.

There are some features Webster would like to see added to Speech Server, including multilanguage support and the ability to override Speech Servers default text-to-speech pronunciations for particular phrases.

From a deployment perspective, Woodforest has a single telephony server that will manage incoming calls through an Intel Corp. Dialogic 48-port telephony card. A Web server runs the banks speech application, and two servers in a load balancing cluster provide speech services. Webster plans to scale the system as needed by adding more ports to the telephony server and more speech servers to the cluster.

Technical Analyst Michael Caton can be reached at

Check out eWEEK.coms VOIP & Telephony Center at for the latest news, views and analysis on voice over IP and telephony.


Submit a Comment

Loading Comments...
Manage your Newsletters: Login   Register My Newsletters

Rocket Fuel