A well-designed IVR system is a great solution to allow customers to
perform quick queries, but a poorly designed IVR system can upset and annoy
customers. We've all experienced a poorly designed IVR system. What
companies need to realize is that customers' patience is limited and they
will quickly 'zero out' to the operator, only to wait another 10 minutes
in queue for a question that the IVR system could have answered. This
common, problematic scenario makes the case for a well-designed IVR system.
In addition to their inherent cost savings, IVR systems can 'offload'
simple, repetitive queries from agents, such as driving directions, account
balances, store hours, etc., leaving agents ample time to address more
complex customer queries.
Until now, IVR systems were limited due to their primary input method:
touch-tone phone pads. With the advent of accurate natural speech
recognition, IVR systems are experiencing a rebirth. LumenVox's Speech
Driven Information System (SDIS) is a natural speech recognition app-gen
wrapped around the company's Speech Recognition Engine and designed to
answer your customers' commonly asked questions. SDIS was designed with
relatively nontechnical people in mind, so developing on this platform is a
fairly easy task. It runs on Windows 2000, supports a range of voice
hardware including TAPI modems and Dialogic hardware, and can scale from as
few as one port up to multiple T1 lines. The company even provides a free
one-port version at no charge that can be downloaded from LumenVox's Web
site.
Operational Testing
One of the first things we noticed about SDIS was how quickly and easily
we were able to design a speech-recognition virtual assistant/IVR. SDIS
comes with several examples, which are helpful when learning to use the
product, as well as a five-minute tutorial wizard. We were able to quickly
set up a test speech-rec IVR system in less than five minutes. Using SDIS, a
caller can simply say, 'What is my account balance?' or 'How do I open
an account?' The responses to these questions are very easily developed
within the SDIS application generator screen, which can be tied to a
back-end database for lookups. In the next release, LumenVox will support
C++ calls to add more flexibility and to control the course of the call. For
example, if an account is past due, the program can send the call to the 'make
payment' menu rather than allowing the caller to gain access to other
information.
SDIS has an attractive, clean interface, but it does take some getting
used to. Although it was easy to use once we got the hang of it, its
usability could be improved. For instance, we tried right-clicking on the
elements within the Call Flow View (see Figure 1) in an attempt to rename
it, but right-click functionality was not available. Also, from the Call
View, we thought we could create a 'goto' statement (connecting
elements) simply by drawing a line between the two elements, but this
feature was unavailable. Two other key usability features that were
unavailable include 'drag and drop' and 'copy/paste' capabilities.
Adding new phrases is accomplished by typing the phrases into SDIS. A
user need not record the phrases into the system since the speech
recognition engine is speaker independent. Instead, the system uses phonetic
spelling to match the typed words to spoken speech. You can even organize
several phrases to be matched to a single action. For example, the phrases
'Tell me my balance,' 'What is my balance?' and 'How much money do
I have left?' can all result in the account balance being spoken. SDIS
also lets you record your voice prompts from within the application. You can
record a new prompt or link to a previously recorded prompt in the Library.
You can also type a 'transcript' within the record window so you can
simply read the transcript into the microphone. You can also play back the
recording and re-record it to ensure it's exactly as you wanted it.
We particularly liked that SDIS supports using a sound card for
development and testing. When we made our test calls, for ease of testing we
used a sound card with a headset instead of a telephony board. In any event,
when we tried some of the samples, we spoke our responses into the
microphone and the natural speech recognition engine was quite accurate.
SDIS supports barge-in, so there is no need to wait for a prompt to finish
before making a request. In addition, it supports an 'out-of-vocabulary'
filter that weeds out extraneous utterances such as 'um' and 'ah.'
TMC' Labs liked the SDIS Call Engine GUI, which displays the current
status of the call along with the words recognized and the confidence score
in the recognition. It also displays the last concept and the last module
recognized. This is useful for debugging your application. Another useful
debug utility is the Call Flow application, which shows the call flow,
organized in a color-coded hierarchal layout. The module name and confidence
score are displayed throughout the call flow. In addition, an audio button
is displayed next to each module with the caller's corresponding recorded
phrase. By pressing the appropriate button, a developer can determine why
the confidence score was low or why a spoken phrase was recognized
improperly.
The documentation was exclusively comprised of online help files, but
this was not necessarily a drawback. Most developers prefer using online
help files when looking up a particular API command or working within a
development program. For non-techies who absolutely require a printed
manual, the help files can be sent to a printer.
Room For Improvement
We noticed that we couldn't cut and paste or drag and drop elements
from within the 'Question' screen, a feature we think would be useful.
Also, within the Call Flow View, we couldn't select individual elements
and duplicate them by using the copy/paste method. This would be helpful
when menus are very similar and require only minor modifications. A MRU
(Most Recently Used) list of recent projects in chronological order in the
File menu would be beneficial, as well.
The Call Flow View interface only shows the arrows (goto commands) coming
from or going to an individual screen element that you clicked on. We would
like to see all of the connecting arrows displayed at all times, or at least
the ability to turn the arrows on and off at will if the display becomes too
cluttered.
Conclusion
LumenVox's Speech Driven Information System is a powerful, yet
easy-to-use speech-recognition IVR development platform. TMC' Labs was
very impressed with SDIS's feature-set, including barge-in and very good
natural speech recognition engine. At $250 per port for SDIS and $250 per
port for the Speech Recognition Engine, the product is very competitively
priced. With LumenVox's free one-port version available for download on
the company's Web site, we recommend that companies that currently use an
IVR explore LumenVox's offering.
[ Return To The
June 2002 Table Of Contents ]
|