A voice verifier for face/voice based person verification system

A person verification system based on voice and facial images has been developed within CSIRO Telecommunications and Industrial Physics, Australia, for use in low-to-medium security systems. It provides a unique ID, which is non-intrusive, fast, and has no need for memorising passwords. A stand-alone version of the voice verifier has an error rate of less than 8%, while the face verifier has an error rate of less than 5%. By combining the two modules, an error rate of less than 1% is achieved. This paper describes in detail the method and some of the important practical issues in the implementation of the voice verifier. It also addresses the issue of decision making if the two sub-systems produce contradictory results.