ServicenavigationHauptnavigationTrailKarteikarten


Research unit
EU RFP
Project number
97.0494-2
Project title
PICASSO: Pioneering caller authentication for secure service operation

Texts for this project

 GermanFrenchItalianEnglish
Key words
-
-
-
Anzeigen
Alternative project number
-
-
-
Anzeigen
Research programs
-
-
-
Anzeigen
Short description
-
-
-
Anzeigen
Further information
-
-
-
Anzeigen
Partners and International Organizations
-
-
-
Anzeigen
Abstract
-
-
-
Anzeigen
References in databases
-
-
-
Anzeigen

Inserted texts


CategoryText
Key words
(English)
Speaker verification; synchronous alignment; incremental enrollment; customized password; banking an telecom applications
Alternative project number
(English)
EU project number: LE4-8369
Research programs
(English)
EU-programme: 4. Frame Research Programme - 1.1 Information technologies
Short description
(English)
See abstract
Further information
(English)
Full name of research-institution/enterprise:
Institut dalle Molle d'intelligence artificielle perceptive IDIAP
Research Institute
Partners and International Organizations
(English)
IRISA (F), KTH (S), KUN (NL), ENST (F), UBS-Ubilab (CH), IDIAP (CH)
Abstract
(English)
Speaker verification concerns the access control and user authentication through voice prints, more specifically verifying the claimed identity of a user through his/her voice. In this framework, the present PICASSO project was about the development, testing, and improvement of state-of-the-art speaker verification approaches, in the framework of telecom (through our industrial partners Swisscom. CH, and KPN, NL) and banking (through Ubilab, CH, and Fortis, NL) applications. More specifically, PICASSO had two principal objectives:
1. Formal evaluation of the performance, and resulting acceptability by the users, of a state-of -the- art system on real task scenarios (as referred above). The initial system. resulting from a previous EC project (CA VE) in which illIAP significantly contributed, has indeed been shown yielding state-of-the-art performance in the framework of NIST (National Institute of Stamard and Technology, USA) international evaluation campaigns.
2. Development of a demonstration and real time prototype system. which could be exploited by Vocalis, an English speech vendor also partner of the project.
During the course of PICASSO, several research directions were investigated by illIAP, resulting in significant improvements of the above state-of-the-art system, including:
1. Incremental enrolment: allowing online/unsupervised adaptation of the parameters of each customer model, resulting in incremental adaptation and improvement of the system.
2. Synchronous alignement: allowing to use speaker independent models to boost and improve speaker specific models.
3. Improvement of the score normalization (hypothesis testing) and decision strategy. More recently,
this was also further improved by using new machine learning approaches, such as Support Vector machine.
4. User customized password, allowing each user to choose his/her own password, simply by pronouncing it a few times.
All this resulted in a re-usable software package which is still in use today at illIAP.

References in databases
(English)
Swiss Database: Euro-DB of the
State Secretariat for Education and Research
Hallwylstrasse 4
CH-3003 Berne, Switzerland
Tel. +41 31 322 74 82
Swiss Project-Number: 97.0494-2