"A Multilingual Speaker Verification System: Architecture and Performance Evaluation"


Javier Caminero (Telefonica I+D) 

Joaquin Gonzalez-Rodriguez (ATVS-DIAC-Universidad Politecnica de Madrid)

Javier Ortega-Garcia (ATVS-DIAC-Universidad Politecnica de Madrid)

Daniel Tapias (Telefonica Moviles Espaņa)

Pedro M. Ruz (Telefonica Moviles Espaņa)

Mercedes Sola (Infomatica El Corte Ingles)


EP1: Evaluation


In this contribution we present a multilingual secure access front-end that checks the identity of the user of a service through the mobile, PSTN and the IP networks (G.723, G.729). Our system prototype is based on speech recognition and speaker verification technologies and it uses a decision mechanism to combine the outputs of both modules. The main objective of the system is to increase the services access security with no increase of the service complexity. The system initially works in six European Languages (Spanish, English, French, Catalan, Galician and Basque) even though the system architecture easily allows the addition of new languages. The system has been developed through a EC funded project called SAFE (Secure-Access Front End, IST-1999-20959).


Multilingual, Speaker-Verification, Evaluation, Architecture

Full Paper