By Stephen Ballantyne
Friday 9th May 2003
|Text too small?|
It's listed on the ASX and it has many customers in Australia but the research and development arm is still based in New Zealand where it has a staff of 18. According to Tom Risbrook, VeCommerce's New Zealand country manager, "we're not a division of someone else all we do is speech recognition."
The VeCommerce solution probably most familiar, at least to Aucklanders, is Ivy, the system the company devised for Auckland Co-op Taxis.
Call Co-op for a cab and Ivy will look up your telephone number in its database of previous customers, tell you what it thinks your address is and ask if you're ready for a taxi. If you say you are, it will automatically send one around.
It's about the simplest imaginable voice-based system (although there are some subtleties that most users won't encounter for example, if you phone back within 20 minutes Ivy connects you to a human operator who will tell you how long your cab will take to arrive) but it works very well. Even at peak traffic times calls are answered immediately and the number of staff has been seriously reduced.
"We have tended to focus on five marketplaces government, finance, travel, wagering and utilities," Mr Risbrook said. "From a business perspective, VeCommerce is mainly interested in routine high-volume transactions, where clients are able to complete the transaction by talking to a computer in a natural conversational way.
"On a technical level, we have a solution with many elements a phonetic model, which is concerned with understanding how Kiwis talk; another is the application itself and how it sounds.
"We're mainly interested in speaker-independent solutions, which are different from the desktop applications you may be familiar with, where you train your computer to recognise your particular voice. Our speaker-independent solutions attempt to cope with anybody in New Zealand, where there's a huge variety of accents."
The system analyses what is said on the fly, comparing expected results with individual vagaries of pronunciation to determine how a particular user's version of various phonemes matches standard local pronunciation.
"We might ask you to say your account number; our recogniser will generate a confidence score based on what it recognises.
"For example, we installed a system called Susan for Work and Income [VeCommerce's solutions are given names that reflect the voices they use], which is capable of giving you your benefit details. The client base consists of people in lower socioeconomic groups people on benefits.
"That includes a lot of people from minority groups with accents that differ from standard New Zealand English and the system required a lot of tuning. When we first went live our recognition rate was only 50%; after tuning, we're now up to 98% recognition."
According to Mr Risbrook, automated call management systems that involve pressing keys on a touch-tone phone to select from menus of choices deserve the reputation they have earned for long-winded inflexibility. "With speech, it's easy to start with an open question such as 'Welcome to ABC Co, what would you like to do?' and then channel down very specifically to what you want to do.
"You don't need to know that there are four or five other options you just say what you want and the system compares that with what it knows you might be after and moves you on to the appropriate next stage or connects you to a human operator if necessary.
"As far as take-up rates and user satisfaction are concerned, we're finding that people tend to quite like it. Although automation is usually seen as a cost-cutting exercise, speech recognition-based systems are actually popular. Often people don't want to talk to another human why bother, when what we really want is a taxi, not a conversation?
"The internet is seen as the most economical self-service system but not everyone has a computer or wants to use one to order things. We work in the same way as the net, connecting to the same host database and applying the same business rules but using the phone rather than a browser. And there's a lot more phones in the world than computers.
"Our value proposition is in reducing the number of staff needed for call-centre operations; currently about 4% of contact centres in Australia and New Zealand are using speech recognition and the figure is expected to rise to 17% by the end of the year."
VeCommerce expects to do well out of this and not only in Australia and New Zealand. It has also installed systems in the UK, notably for bookmakers Ladbrokes and pools company Littlewoods.
VeCommerce's expertise isn't so much in the underlying technology as in the way it puts together total systems. The mechanics of speech recognition were worked out years ago by US university researchers and if such systems seem to be working better these days than they did in the past it's mainly because the hardware is much faster than it used to be fast enough to analyse speech as quickly as humans do.
But no computer has any real experience of life; the context of taxis, social welfare systems, fast-food ordering and bet-placing has to be provided by real people.
Mr Risbrook demonstrated Kelly, a take-away ordering system soon to be offered by a large local fast-food chain. Kelly appears to work very well, thanks to VeCommerce's careful circumscription of the system's field of discourse and matching of telephone manner to customer expectations.
Kelly sounds like the sort of kid you'd expect to find working in a takeaway joint, but even though she's unmistakably assembled from pre-recorded fragments, the temptation to speak to her as if she is a person is strong. Nor is it wasted including "please" and "thank you" in a conversation with Kelly provides useful markers that the system is sophisticated enough to interpret.
The system is satisfying enough for customers that some try to extend the interaction Mr Risbrook has a tape of comments recorded from customers of an Australian insurance company VeCommerce constructed a system for mostly thanking the system for being helpful and efficient and one asking "are you real?"
No comments yet
NZ dollar headed for 1.3% weekly gain on expectations of a Fed rate cut
RBNZ knock-back gives Resolution chance to low-ball AMP - Jarden
Rail hubs may not boost Napier Port log trade
O'Connor looks to overhaul Biosecurity Act, improve animal tracing
Denton Morrell undefended at liquidation hearing
Contact steam to heat Norske Skog pellet business secured
Air NZ to amend booking engine after lawyer’s complaint
Ross McEwan to take helm at NAB
KPMG says bank capital proposals will wreck havoc on dairy farmers
Mild weather saps Vector's June-qtr volumes