International Working Group on Data Protection in Telecommunications Common Position on Speaker Recognition and
Voice Analysis Technology in telecommunications adopted at the 25th Meeting of the Working Group on 29 April 1999 in Norway Among the currently developed biometrical identification methods, speaker recognition is probably the most advanced and of particular relevance to telecommunications. Speaker recognition is a method to analyse features of a person's voice to:
In all cases a person's voice is measured and compared to a previously recorded and stored digital template or voiceprint of his/her voice. Best results in recognising persons, in terms of failure rates, are obtained if the same words are used for input and for the template (text dependent systems). Think of a predetermined password or ID. When entered, this is matched to a stored voiceprint. In other systems speakers are prompted to repeat randomly selected words, which are being matched to the template (text prompted systems). An advantage is that the system cannot be misled by impostors who use voice samples recorded on tape. Finally, in text independent systems a person is asked to talk and his utterances are matched with the stored templates, containing completely different words. This situation offers much more contingency, and hence the matching is more difficult, in particular if background noise is present or noisy telephone lines are used. On the other hand the potential of these systems is high: combined with a large database of voice templates, a text independent systems enables identification of many different persons in many circumstances. Speaker recognition can be used for identification and authentication for both access to the network and equipment and access to the services delivered over the network. Obviously telecom operators perceive an interest in improved voice identification and authentication for various purposes, for instance fighting telecommunications fraud or marketing of new features and services. As for services delivered by means of the telecommunication networks, identification of customers is increasingly seen as an important for making on-line decisions on the way an individual is treated. It should be noted that, unlike most other biometrical identification methods, speaker recognition does not need a new infrastructure, but can be integrated in the existing telecommunications networks. The use of speaker recognition is still restricted to dedicated applications. The cost level of this technology is, however, expected to decline rapidly, while the quality of the systems is continuously improving. Mass applications can be expected in the near future. Data Protection and Privacy Commissioners have stated on other occasions that anonymous means to access telecommunication networks and anonymous means of payment are two essential elements for true online anonymity. The International Working Group is especially concerned about the risk that these techniques may be installed and used in telecommunication networks without any knowledge of the users or any means to avoid this phenomenon. Recommendations
|
||||
|