Talking computers will soon replace recorded messages and human operators in a range of telephone services, such as travel information and home banking. The computer will ask callers what they want to know, retrieve the information from its database and convert it to speech.
Interactive telephone services already exist, but they require people to use touch-tone dialling ('Press one for yes, two for no'), or to respond with answers from a limited list of words that the computer can recognise. The new services will allow users to talk naturally over the telephone to get information, pay bills or buy a suit.
Staff of British Airways at Heathrow are currently testing a system that allows them to ring up and speak to a computer to extract flight information from its database and make bookings. The system, known as Callserver, was developed by Vocalis, a company specialising in speech recognition that was set up last May after a management buyout from the computer services company Logica. Vocalis says Callserver could be used for automated receptionist systems, transaction processing, voice messaging and information services.
The Swedish telephone operator Telia is about to begin a trial of Callserver, which it plans to use for directory inquiries, accounts payments and other services.
BT is developing a number of interactive services in which computers recognise complex spoken commands such as names, addresses and strings of numbers, and respond with speech, allowing a sophisticated dialogue to be held between a computer and a telephone caller.
BT Laboratories, BT's research arm, recently demonstrated its prototype interactive telephone service, a bill payment system. Customers ringing up are greeted by a recording that asks for name and identification number. Once their bona fides are established, the computer will look up account details and tell callers how much they owe.
Although it is slightly Dalek-like, the computer-generated speech is intended to sound the same as the voice on the recording. The computer will then respond to spoken instructions on how much callers want to pay and the method of payment, reading the details back to ensure they are correct.
BT has begun to use the technology to automate its directory inquiry service, so that callers do not speak to a human operator at all. It is also running trials of the system, known as Call Centre, at a large commercial customer in London.
Chris Southcott, at BT's Speech Applications Division, says many companies are now trying to get their customers to deal with them on the phone rather than using the post, because it is faster. Companies with large numbers of operators handling telephone queries will be able to transfer routine calls over to the computer.
'The system could take customer orders, accept bill payments or give out timetable information. This will free staff for complex calls, such as dealing with complaints,' Mr Southcott says.
The company is also running trials of a sophisticated answering service, Call Minder, in London. Subscribers can have their phones answered by a computer, which will ask the caller's name, message and return phone number.
Speech-recognition systems that turn a user's speech into text are already commercially available. IBM launched a speech-operated word processor earlier this year. The problem with many such systems is that they need to be 'trained' to understand an individual's voice, and users have to pause between words. This makes them unsuitable for telephone services, which must deal with thousands of different voices.
Mr Southcott says the BT system can understand 90 per cent of voices without difficulty, and only has problems distinguishing some words spoken in an extreme regional accent.
Banking, insurance and travel companies are showing great interest in the new interactive telephone services, not least because they are cost-effective. 'The cost is typically pounds 2,000 to pounds 3,000 per line, and the business payback can be as short as six months,' says Jeremy Peckham, managing director of Vocalis.
The systems are not yet sophisticated enough to replace human operators, but they are able to handle lower-grade inquiries. 'There's nothing more frustrating than waiting in a long queue when all you want is the answer to a simple query,' Mr Peckham says.