A Speech Biometrics System with Multi- Grained Speaker Modeling
Jií Navrátil, Upendra V. Chaudhari, Stephane Herman Maes · 2000
The paper describes a system for voice-based personal authentication in a conceptual framework of conversational speech biometrics relying on two sources of authentication - the acoustic voice-print and the user knowledge. Several technologies are closely integrated: speaker recognition and speech recognition as well as natural language understanding and dialog management. The first part of this paper details on the algorithms for the acoustic voice-print modeling and recognition and presents experimental results on two databases with telephony speech. The second part describes the integration of acoustic voice-print recognition and speech recognition in a prototype that allows for natural language input and dynamic authentication scenaria in the telephony environment, together with an experimental evaluation carried out on a speech-biometrics database.