VOICE BIOMETRICS WITHIN THE FAMILY: TRUST, PRIVACY AND PERSONALISATION

摘要

Driven by an increasing need for personalising and protecting access to voice services, the France Telecom R&D speaker recognition system has been used as a framework for experimenting with voices in a family context. With the aim of evaluating this task, 33 families were recruited. Particular attention was given to 2 main scenarios: a name-based and a common sentence-based scenario. In each case, members of the family pronounce their complete name or a common sentence respectively. Moreover, this paper presents a database collection and first experiments for family voices. The results of this preliminary study show the particular difficulties of speaker recognition within the family, depending on the scenario, the genre and age of the speaker, and the physiological nature of the impersonation.