A UNIVERSITY of Queensland research team headed by a Hong Kong-born professor has developed an automatic teller machine that identifies customers' voices. The machine, which has a current success rate of 999 people out of 1000, recognises each customer's distinctive voice signature - a breakthrough the 10-member research team says will give added security to sensitive transactions. ''The main one is additional security for credit-card transactions by telephones,'' said Professor Tsoi Ah-chung, Professor of Electrical Engineering with the electrical and computer engineering department. Customers applying for credit cards could provide voice samples to be stored on computer then, when they wanted to withdraw money, would be asked to repeat randomly chosen words or phrases. ''Our prototype system can be used to conduct transactions, such as shopping by telephone, authorising long-distance phone calls and seeking details of bank accounts by telephone - in all cases, with greater security using speaker verification,'' he said. Prof Tsoi said the prototype looked like a workstation, but was made to look like an ATM two months ago for a demonstration. It had been tested by verifying the identity of students entering the department's PC laboratory, and would be finished by December. The researchers are seeking funding to develop it further for export. ''What we really have is a core system which does speaker verification. If you are a bank and want to use it for ATMs, we can customise it. If you are in building security, we can customise it for that,'' Prof Tsoi said. ''I have been talking to potential investors with the university people, for instance banks and ATM manufacturers. This certainly has large export potential; someone told me there are more than 10 million ATMs in the world.'' Prof Tsoi said the A$1.7 million (about HK$8.5 million) project was one of five funded through an A$87 million syndicated research and development scheme, the university's largest private research contract. It was funded by Bankers' Trust Australia, which owned the technology jointly with the university. The system was based on new technology, known as artificial neural networks (ANN). These were electrical circuits built to reflect some of the brain's learning processes. Their ability to recognise patterns were helping to overcome problems, such as speech recognition and computer vision. Previous speaker verification systems relied on conventional computing and did not perform well enough for complex uses. Handwritten signatures, although useful, were inadequate in the electronic age, Prof Tsoi said. ''A preferable system for a range of transactions today would be based on a person's voice imprint,'' he said. Prof Tsoi graduated from Hong Kong Polytechnic in 1969. He did postgraduate studies in Britain, where he worked for five years. He then worked in New Zealand for seven years before moving to Australia. He returns to Hong Kong often to visit relatives. Prof Tsoi, who has worked in electronic engineering for 20 years, said he came to the University of Queensland three years ago because it had the largest group of researchers working on ANNs in the Southern Hemisphere. His team developed its ATM prototype in only 21/2 years.