Can we talk?

Communications News, April, 1998 by Derrek Schartz

Coca-Cola New Zealand improves its call handling with a voice recognition system that accommodates multiple pronunciations.

At Coca-Cola New Zealand Corp., a distributor in Auckland with more than 600 employees, a speech recognition system is solving a call volume problem that used to leave both inbound and internal callers in perpetual telephone limbo.

Today, anyone who calls the company is greeted with the message, "Please say the full name the party you wish to talk to." Say "Tom Jones"--or even the name of the department you I re looking for--and your call is instantly routed to the appropriate extension.

No customer needs to hang on the line waiting for an available operator, no employee needs to rifle through a long and sometimes outdated phone list to locate a colleague, and no caller needs to spell a name on a telephone keypad. The system also has eliminated difficulty in reaching the proper party after hours, on weekends, or at other times when an operator is not available.

The computer telephony product Coca-Cola uses is Pure ReQuest, a speech-driven automated attendant from Pure-Speech, Inc. of Cambridge, Mass. Running on Pentium processors with Dialogic cards on the Windows NT platform, it allows callers to speak naturally and accommodates multiple pronunciations for each person, department, product or service name.

The automated attendant as configured and installed at Coca-Cola New Zealand in just 10 days last fall, and it has achieved recognition accuracy levels greater than 98% since then.

"When we started talking Coca-Cola in September, they were just installing their first voice mail system, and that was going to solve the problem of having operators tied up taking messages," says Geoff Witton, president and CEO of Speech Recognition Systems, Ltd., a New Zealand firm that handled the Coca-Cola implementation and is using the system in its own offices as well. "But it still left hundreds of calls every day that needed either operator assistance or a physical lookup in the company directory just to find the proper extension.

"Pure ReQuest made it possible to handle those calls almost instantly, expediting employee-to-employee communications and ensuring that customers wouldn't get discouraged by long waits," he says. "It also enabled the company to avoid adding more operators and phone lines to handle increased call volume, and it's saving money on toll-free phone bills by reducing wait times."

Touchtone-activated auto attendants that invite callers to spell the name of their party on their telephone keypad are common today, but the speech-activated varieties like the one used by Coca-Cola New Zealand are new arrivals in the computer telephony marketplace. These products listen to the human voice and break it into its phoneme blocks, then compare those tiny units of speech against a preloaded template and look for a match according to a user-defined scale (for instance, 90 out of 100). Matches that meet the required score are routed to the designated extension. Those that fall below the prescribed score are double-checked with a message like "Did you say Susan Smith?"

Coca-Cola New Zealand's new system is an open one that interfaces with any PBX that supports analog stations as well as older electronic key systems. This makes it usable in virtually any setting British or American English or Latin American Spanish, including overseas applications with older phone systems that may lack capabilities like touchtone locators and voice mail that help callers reach their destinations faster.

The system disregards superfluous words and phrases such as "uhh," and it has a "barge-through" feature that allows callers to begin speaking without waiting for the prompt to finish playing. It also allows administrators to quickly add, delete, or edit name pronunciations and update the system in real time, without interruptions in service.

In the case of the Coca-Cola New Zealand installation, Speech Recognition Systems already had made the minor phonetic modifications necessary to enable the system to understand local speech patterns. All that remained was to define the on and off phone tones at Coca-Cola, import the company directory into Pure ReQuest and then adjust the database to ensure that all possible permutations of each name would be recognized (for example, Fred, Freddy, Frederick).

When those tasks were completed, a server loaded with Dialogic telephone interface boards, Dialogic Antares voice and digital sound processing boards carrying PureSpeech's proprietary speech recognition algorithm, and Pure ReQuest software was plugged into a port on Coca-Cola's PBX, and the system was up and running.

To streamline the integration process, PureSpeech uses the technology division of Dallas-based Alliance Systems to assemble a turnkey hardware platform for its customers.