July 1998

NameConnector

Parlance Corporation
200 Boston Avenue, Suite 1500
Medford, MA 02155
Ph: 888-700-6263
Web site: www.nameconnector.com

RATINGS (0-5)

Installation: 5
Documentation: 5
Features: 4
GUI: 5
Overall: A-

The Parlance NameConnector service uses speech recognition to enhance the capabilities of the corporate telephone switch. That is, NameConnector replaces touch-tone as a way to reach corporate personnel. Instead of poking through a touch-tone system, a caller can simply utter a word - a person's name, a department's name, or even a common question - and reach the appropriate extension. Thus, if a caller forgets an extension number, he or she needn't listen to lengthy messages or enter the "first three letters" or negotiate tedious menus. Basically, with NameConnector's easy-to-use interface, and with Parlance's online support, corporations can bring the benefits of speech recognition to any corporate telephone switch.

INSTALLATION/DOCUMENTATION
We received NameConnector as a turnkey system installed on a 266 MHz Pentium II PC with 128 Megs of RAM running under Windows NT 4.0. We connected the computer to our PBX system via two extensions, which could be dialed from either inside or outside the company.

We uploaded a simple tab-delimited text file from a 1.44-MB floppy drive using the Import Text Button on the NameConnector Control Center toolbar. This created a preliminary database so that we could begin testing the system.

The documentation for the product was well organized, thoroughly indexed, and covered just about every topic we needed to know about. Actually, we were able to work without digging through the manual at every step. The system was very easy to use. Also, we were able to take advantage of the very good online help, which was especially useful when we implemented some of the more advanced functionality associated with actual telephony properties.

FEATURES

• Arrives completely installed. Besides installation, Parlance provides ongoing customer support and system monitoring and maintenance.
• Maintains a database of up to 10,000 names complete with pronunciations and alternate pronunciations to increase accuracy in matching.
• Uses tab-delimited text for easy updates from any database.
• Plays back names before transferring the call, giving the caller the chance to exit before reaching the extension.
• Allows callers to reach the "Operator" verbally or by pressing "0."
• Monitors the phone system's lines, and alerts the user if any lines are not functioning.
• Edits the pronunciation of names to maximize matching success.
• Compiles reports about call volume and system accuracy. The report utility lets you review daily, weekly, and monthly results.
• Automatically backs up the service database and files.

OPERATIONAL TESTING
The small database we used was compiled from an Excel workbook with about 50 entries supplied by the vendor. The file contained the names of some of our employees (including nicknames and common mispronunciations) and the corresponding phone extensions. For each item listed, we had room for five alternative names. We didn't explore nearly enough mispronunciations to even approach the system's limits. (We take this as an indication of the system's strength, as opposed to any lack of imagination on our part.)

Although we had little difficulty using an Excel file, we noted that Excel did not come installed on the computer, so we had to do our editing on a different machine. When we tried to edit the text file directly, we encountered some problems formatting the tabs and spacing correctly, so we returned to using the Excel file.

After we imported the text file, we went through a process that allowed us to check the pronunciations of all the names. The program uses some grammar rules, which allow it to approximate the pronunciation of most names. With these rules and the text-to-speech function, the computer can "speak" the name over the multimedia speakers. If a name sounds incorrect, the program can change which parts of the name are stressed, thanks to a built-in utility. This utility worked for all but one of the names in our directory. For some reason, the name Rita Budahazy proved a challenge. In this case, we had to spell the last name as "Buddha Hazy" for the system to pronounce the name correctly. Still, the automatic text-to-speech worked for 98 percent of the names in our database.

The system was set up as one of our extensions so we could dial it from anywhere to see how it worked. When we called the extension, it would play a greeting (which we eventually shortened to save time) and then ask for a name. We had to wait for the prompt before we spoke the name, which seemed a nuisance at first; however, after we discovered how accurate the recognition was, we decided that observing this formality was worthwhile.

We updated the text file a few times to learn more about the system. Each time we imported a file with changes, the program gave us a window with a summary of exactly what had been changed, including which items had been ignored, dropped, added, or updated. The only time we ran into a snag entering a name was with Mary Beth D'Arcangelo. We had used an ASCII symbol that looked like an apostrophe, but wasn't. This caused the D' to be ignored. We quickly updated the file with the proper character and everything was fine.

Another NameConnector feature allows it to dial outside lines in addition to in-house extensions. We had to type in a 9 before the number to get out of our system, but once this was done, the program dialed out just as easily as reaching any internal extension.

ROOM FOR IMPROVEMENT
We found the Parlance NameCon-nector to be a decent system. About the only thing we found ourselves searching for was another way to edit the text file. Thus, we would like to see Parlance include some sort of text editor so users needn't resort to using another computer to make simple changes and corrections. Of course, some users might prefer using another computer, such as users who import information from another database, and who want to use the original database with all its functionality.

We'd also like to see some embedded utilities to facilitate the modification of dialing properties. To change the amount of time between getting an outside line and dialing the rest of the number, we had to change the registry, which may be a problem for some users. Such users might like a utility that would let them define "custom dialing" specifications. Such a utility would allow the computer to use its logic as to how to dial the number, especially when dealing with a home number.

In addition, we noted that if a caller knew the extension of the party he or she wanted to reach, there was no way to simply type in the number to exit the speech recognition protocol. While at first we thought this was standard with the system, we did notice that the caller was able to dial a "0" for the Operator. Any other key press would give us the message, "Invalid key press. Please speak after the tone." If the product were expanded to include the possibility for a caller keying in the extension, Parlance would accommodate either method of reaching an extension, that is, by direct dialing or by going through a directory. Instead of imposing the directory method in all circumstances, the product should give users the choice of either method of reaching someone's extension.

CONCLUSION
The Parlance NameConnector is a quality product that does exactly what it was designed to do. In essence, it makes a corporate telephone system friendlier to the caller while maintaining the accuracy of conventional call routing programs. Even though speech recognition is far from perfect, the system works well enough to make it worthwhile. In our tests, we occasionally had difficulty completing a call. But in every instance, the difficulty was attributable to some mistake on our part. For example, there were times we attempted to use the system over a speakerphone even though we were sitting too far away.

When we spoke into the handset, we were impressed with the ability of the software to route our call to the proper extension. Things were a little more erratic when we used partial names, but with proper maintenance of alternative names in the database, this product could easily perform well enough to save money for the companies that implement it. Essentially, the key question is this: Does the product improve efficiency enough to justify the very small sacrifice in accuracy? The answer is yes.