For most of the broad technology community, speech recognition falls into the cute but not very useful category. The call centre industry, on the other hand, sees it as the tool to free users from the dreaded "press one for the main menu" DTMF systems.
[VIDEO]Dual Tone Multi-Frequency took away the customer experience, says Haydn Faltyn, speech business unit manager for Dimension Data Australia. "The human being is most effective when he can speak out and take text in."
He sees a future where multi-modal interaction sees people using voice commands to surf text displayed on a personal digital assistant (PDA) while being prompted by a soothing female voice, instead of punching numbers into a keypad.
[VIDEO]There is do doubt in his mind that telephones will continue to be the pervasive tool for communication between individuals and companies, even as call centres are integrated with Internet contact points.
"The premise for speech is that the phone is everywhere," he said, speaking at a DiData breakfast on the opening day of the Customer Contact World exhibition. "There are parts of the world, specifically in this region, where phones are being rolled out as brand new technology." While telephones are rudimentary compared to PCs, they are still connected devices.
[VIDEO]Call centres receiving thousands of calls a day have long since realised that any automation that could cut down on call duration would save huge amounts of money. Most call centres implemented front-end systems to cull at least a minimum of customer data before the call was transferred to agents. However, most people were unwilling to use the systems, says Faltyn.
A case in point is Telstra, an Australian telecoms operator for which DiData is about to roll out a system. While the DTMF system provided almost no information, a speech recognition system that forces callers into using it has racked up a satisfaction rating of more than 80%.
Recognition does not need to be totally accurate, argues Faltyn. "Recognition doesn`t have to be 99% to be effective," he says. "It needs to be better than your business case."
Putting together the project
Contrary to popular perception, software development does not comprise the major expense in a big voice project such as Telstra`s. Faltyn says application developers comprised less than 10% of the staff on the Telstra project, with more grammar developers employed, and about twice as many dialogue designers used. That excludes the psychologists who helped create the persona for the voice that interacts with callers.
[VIDEO]The description of that persona, as envisaged in the planning stage, makes for interesting reading. A plump middle-aged female with coloured hair and a love for sport may not be everyone`s dream girl, but was envisaged as comforting when dealing with account enquiries.
Real-life examples of applications are wide and varied. Faltyn says Didata has worked on a system for the Australian taxman for the filing of personal income tax forms by telephone. Australian law recognises voice print identification as legally valid and gives it the same weight as a signature.
Future applications can include web surfing, voice information portals and more mundane directory assistance.
As a rule of thumb, Faltyn estimates a typical new-breed voice project takes between six and 12 months to complete, with small changes in the interface taking up to six weeks and larger structural changes adding 12 weeks to the duration. Because of such tuning, up to 30% of the total project cost can be incurred after the system is launched for public use.
Share