|
Advanced speech technologies offer enterprises new self-service
approaches, with speech recognition providing a direct link between
customer needs and business activities. Just as Web portals have emerged
for users equipped with browsers, we expect voice portals, enabled through
advanced speech technologies, to dramatically enhance customer access to
automated services and provide companies with substantial savings and
increased competitive edge.
This access can be provided for 830 million wired phones and 470
million mobile phones, and leverage nature's browser built into every
human being. Customers can say what they want directly without having to
navigate through multi-level menus. More importantly, these technologies
can open up new services for telephone users, that would be prohibitively
cumbersome to implement using multi-level menus. The key technologies
(large vocabulary recognition and natural language understanding, speaker
verification, and advanced text-to-speech) behind the following scenarios
are here now. Which would you prefer?
SCENARIO 1
Customer: I want to apply for a new home mortgage.
Machine: How much do you want?
Customer: As much as I can get!
Machine: You will have to be more specific. What amount to the
closest $1,000 would you like?
Customer: $150,000
Machine: What kind of mortgage do you want?
Customer: Darned if I know.
Machine: Let me explain the options.
SCENARIO 2
Machine: For account updates, press 1, to pay a bill, press 2; to
report a stolen card, press 3; .... for mortgage applications, press 23.
Customer: Maybe I can get service somewhere else?
If this was your choice, then your preference would be obvious. You're
only human!
SAYING WHAT YOU MEAN
Today's large vocabulary recognition technology can recognize thousands of
words spoken naturally by the caller with a broad range of accents and in
a speaker-independent fashion (i.e., no training), not only when spoken as
distinct words, but also in continuous speech delivery (a much more
difficult challenge). Natural language understanding derives the meaning
of what is being asked -- even more challenging. Large vocabulary
recognition allows customers to speak their requests, providing them with
an easier, more user-friendly, personalized interface. Large vocabulary
recognition algorithms have made it possible to deploy solutions that
automate significant numbers of customer transactions using spoken input.
They can say what they want directly and provide multiple pieces of
information in a single phrase, thereby reducing the need for multi-level
menus and simplifying the navigation process. The ability of the solution
to recognize natural numbers, money amounts, dates, and times in a range
of formats, can further reduce transaction time. Good examples of large
vocabulary recognition are the automated stock quote/trading systems
deployed by major financial institutions.
Callers get higher levels of satisfaction because they are able to
complete their task using a more natural method of communication.
Institutions can significantly off-load calls from agents, and can offer
unique services by automating revenue-generating calls that are too
complex for touch-tone input. For one major firm, a large vocabulary
recognition system has cut in half the number of calls taken by human
agents. For another, the cost per call was reduced to less than half when
compared to the same function performed by agents. The systems paid for
themselves in months.
GETTING THE RIGHT ANSWER
Large vocabulary recognition is just one side of the solution. There is
also a need to get the appropriate information to the customer.
Text-to-speech allows the system to convert text data obtained from a
database into synthesized speech for the caller to hear. In most cases,
data that is spoken back to the caller is prerecorded and digitized for
the most natural-sounding speech. This is complemented by the ability to
handle very large or frequently changing database information for which
pre-recording would be impractical. Very importantly, the information can
be delivered in a consistent fashion for every caller (as in the
description of mortgage options available).
WHO ARE YOU?
Voice verification addresses the security needs of the high volume call
processing marketplace. As the frequency and value of telephone-based
commercial transactions increase, the need for security and protection
becomes critical. Voice verification complements existing means of
authentication such as touch-tone-based passwords or PINs, and can
significantly reduce an institution's risk of fraud.
Voice verification works on any standard phone and can drastically
enhance caller automation speeds, while increasing the efficiency and
accuracy of interactive voice response (IVR) applications. The IVR
application prompts the caller to "voice verify" their identity
each time they wish to access secure information and services. The spoken
reply is packaged and sent to voice verification engine, which compares
live speech samples against a stored voice print template (a pre-recorded
sample of the user-selected password). The verification engine then sends
a message back to the IVR application, indicating whether access should be
approved or denied.
STATE-OF-THE-ART PLATFORM
A state-of-the-art speech processing platform integrates these
technologies into an IVR and computer-telephony integration (CTI) system.
It includes advanced system software run on high performance processors
with industry standard components to deliver the benefits of an open
architecture. This approach provides a scalable, cost-effective
implementation, which can be expanded to meet the individual requirements
of a particular recognition task or advanced speech processing algorithm.
It can run multiple recognition, verification, and text-to-speech
algorithms simultaneously, for example, to leverage the optimum technology
for a particular linguistic group. The system dynamically allocates the
advanced speech processing (e.g., recognition) resources to the
multiplicity of phone calls active in an application.
WHAT'S IN IT FOR ME?
For customers, IVR systems are convenient, easy-to-use, and non-invasive.
Customers can accomplish more in less time while using a more natural
method of communication. With earphones and microphones, they can be used
safely while speeding down the expressway. People can make routine
inquiries and enter information directly without the need for service
agents. Information prompts and other messages are spoken to the caller in
a clear, prerecorded voice. The customer can decide which language to be
used for transactions. Commands may be given to the system either by
touch-tone or speaker-independent word recognition, which can accept
certain key words as input, regardless of who is speaking.
For enterprises, IVR systems provide greater automation and
personalization of self-service tasks with a higher customer acceptance
rate. They allow customer service agents to focus on value-added
interactions. The chances of a caller receiving a busy signal and the
number of abandoned calls are both greatly reduced because many calls can
be handled at the same time. Service can be provided twenty-four hours a
day, seven days a week. For example, enterprise experience is that
transaction costs via IVR are approaching Internet costs and are much
lower than using live agents. Speaker verification can enhance security,
and dramatically reduce unauthorized access to secure information and
services. Voice print passwords can be selected and enrolled by the
customer, requiring less administrative overhead. Overall, IVR systems can
be a significant contributor to strengthening of customer relationships.
IVR VALUE IS IN THE APPLICATIONS
Organizations all over the world including banks, insurance companies,
funds, and brokerage houses, airline carriers, institutions of higher
education, government agencies, and telephone and utility companies are
taking advantage of IVR technology. For example, typical customer
transactions automated with IVR by financial service firms can go well
beyond mortgage applications and include:
- Account inquiry and management.
- Card activation and credit authorization.
- Transactions (funds transfer, loan payments, credit card payments,
stop payments).
- Retail services (address changes, loan payment calculator, loan
application faxback, merchant check verification, lost card
reporting).
- Automated customer service (interest rate information, rate
offerings, ATM and branch locator, voice messaging).
- Bill payment (schedule, modification, status).
- Auotes, trades and confirmation, and audit.
IVR can also be used to automate administrative tasks such as ordering
supplies, employee benefits enrollment, and more.
Two major challenges exist. An IVR system has to make use of an
organization's information and processing resources. By leveraging vendor
experience in IVR applications and through development of standards such
as VXML, it will become easier to quickly develop applications and
services that interface to enterprise, customer, and product database.
VXML documents will define a set of dialogs that define the service
interaction using advanced speech technologies.
The other major challenge is related to application design. While the
technology is awesome, computers are not human when it comes to
understanding what the customer intends through his or her words.
Designing an IVR application that meets its objectives over 95 percent of
the time requires a good understanding of human behavior in the context of
the application. Vendors with experience in a broad range of application
environments and industries can leverage this brain trust to develop new
applications.
Leveraging advanced speech technologies to create person friendly,
differentiated information, and transaction voice portals can result in a
win-win for customers and enterprises alike.
Tony Rybczynski is director of strategic marketing and technologies
for Nortel Networks' Enterprise Solutions unit. For more information,
visit the company's Web site at www.nortelnetworks.com.
E-mail questions or comments to tonyryb@nortelnetworks.com.
[ Return
To The March 2001 Table Of Contents ]
|