TMCnet - World's Largest Communications and Technology Community

Tracey Schelmetic

[October 27, 2004]

What Does It Take to Implement an ASR Application?

By Tracey E. Schelmetic, Editorial Director, CUSTOMER INTER@CTION Solutions

A special session at Aspect Communications Corp.s eBIZ 2004 conference, presented by Frank Krzywda, Aspect; Paul Wirtz, Manager of Aspect Professional Services; and Mary Arndt, Senior Applications Engineer for SpeechWorks Solutions, a division of ScanSoft.

Its time for you to put speech into your business. Go ahead what are you waiting for? If youre like most companies, youre waiting for someone to tell you where to begin. Speech has become easier to implement lately due to technological advances, but lets face it its never going to be like buying produce.

Were now seeing the genesis of a new breed of professional services specifically designed to help companies through the somewhat overwhelming process of speech-enabling their businesses. One such group comes from a partnership between Aspect ( and the SpeechWorks division of ScanSoft (

The process of implementing speech into your business actually starts long before the kickoff meeting to implement the actual product.

Companies implementing speech technologies into their enterprises need to ask, and accurately answer, some questions of themselves:

  • How do I know speech is a good fit for my business?

  • Which apps will deliver the best results?

  • Whats the ROI going to be?

  • What implementation approach makes the best sense for our business?

  • How do I make sure the app actually gets utilized?

Pre-launch activities need to include branding strategies, helping involve internal stakeholders in the process, and targeting the message depending on customer profiles. Only until these factors are determined should companies proceed with the steps that ultimately lead to a launch.

Aspect Consulting Methodology (ACM) was launched to help companies manage the sometimes overwhelming process of implementing speech. ACM uses pre-defined phases, procedures and standards to ensure consistency. They lay down best practices; identify and help customers avoid known causes of failure; provide a feedback loop to facilitate improvement of the process; and reinforce points of accountability. To help accomplish these goals, dedicated point people at ScanSoft work with designated point people at Aspect to ensure customers are seeing a unified face as they move ahead with their speech implementations.

The five phases of the Aspect and SpeechWorks consulting methodology include:

  • Project implementation;

  • Define requirements; Project planning;

  • Project implementation; and

  • Project closure.

One of the first requirements of professional services for speech implementations is to make sure that your business goals work together. Having ones cake and eating it too in the world of speech would be exemplified by a company that wishes to implement speech to reduce costs, yet insists that callers be immediately routed to a live agent at any time merely by speaking the word agent. Are your expectations too high? Are they mutually exclusive?

Next, the Aspect/ScanSoft team will identify where your sources of information are: where is the information to be delivered coming from? CRM systems? Your Web site? Multiple databases? The IVR? Contact management software?

Next, you are assisted in creating whats known as caller archetypes. Companies are encouraged to take representations from their regular caller population by humanizing their caller profile data. These resulting characters are then matched up against the proposed speech system. This will help a company understand how their existing customers are likely to react to the new speech system.

Now that you know what you want and whether it fits your business and your budget, the next phase involves user interface design. Call flow, the personality of the system, the text of the prompts and the voice talent are all determined at this point. Determining the personality of the system, obviously, depends on your customer. Youll never use the same voice approach for a teen video game manufacturer, for instance, that you would for a high-end financial services company. Different demographics of customers will obviously respond to vastly different voices, audio and speech patterns. Youd never acknowledge a high-end stock transaction with the phrase Right on!, whereas this may be quite suitable for the aforementioned video game retailer.

This sounds fun and intriguing, yes but its not easy. Behind these voice branding and persona creation processes lies a heap of social psychology research. If a customer responds negatively to the persona, that customer is likely to get off the phone quickly, and never return.

The subsequent phase involves the practical, work-horse part of the process: designing the prompts. With a touch-tone IVR, this would be the process of designing the menu tree. Behind this process is the backend database integration (where are the responses coming from?) and the basic telephone/PBX integration.

One of the final, but most critical, phases is usability testing. Once the system is developed and nearly functional, but before it goes live, it must be tested rigorously. (This can be done in Aspects lab, or in the customers lab). Testers are asked to work through the whole system, speaking requests and problems, and responding to the information supplied. Debugging based on these results follows, and the project is then ready for deployment.

Deployment generally goes in three stages: pilot testing (with just a small set of customers put into the system each day); partial deployment (perhaps 20% of your customers are routed into the new speech system); then full deployment, in which 100 percent of incoming callers are routed through the system.

Finally, post-deployment, comes tuning. A company needs to measure the results of the system, identifying and fixing problems, and then measuring success against pre-defined success metrics. Never a static process, all speech applications, no matter how successful, require monitoring over time.

Obviously, this process doesnt happen in days. A well-designed process requires an experienced team to make sure your customers get the most out of the system, without even perceiving that the system actually exists. To them, its just a pleasant way of doing business.

In the end, a good speech system, well implemented, is like punctuation: if you dont notice it, its well done.

Speech Recognition

Technology Marketing Corporation

2 Trap Falls Road Suite 106, Shelton, CT 06484 USA
Ph: +1-203-852-6800, 800-243-6002

General comments:
Comments about this site:


© 2021 Technology Marketing Corporation. All rights reserved | Privacy Policy