Name : Jean-François Bonastre

Institution : University of Avignon – Inst. Univ. France (IUF)


Jean-François Bonastre obtained his Ph.D. degree in 1994 in automatic speaker identification using phonetic-based knowledge. He is full professor at the University of Avignon and a member of the Institut Universitaire de France. He has been vice president of University of Avignon since December 2008.

As a member of the Natural Language Processing Group, he developed his research in speaker characterization and recognition using phonetic, statistic and prosodic information, while teaching and lecturing on various subjects covering computer science, speech processing, audio signal classification and indexing, and biometry.

From 2001 to 2004, he was the chairman of AFCP, the French-Speaking Speech Communication Association (currently a regional branch of ISCA). He was also the chair of the ISCA Speech and Language Characterization SIG for two years and he joined the board of ISCA in 2005.

He has been vice president of ISCA since 2007. He is a member of the IEEE Speech and Language Technical Committee and IEEE Senior Member.


Publications : (2 maximum)

Fauve, B. G. B., Matrouf, D., Scheffer N., Bonastre J.-F., Mason, J. S. D., “State-of-the-art performance in text-independent speaker verification through open-source software,” IEEE Trans. Audio, Speech and Language Processing, vol. 15, no. 7, pp. 1960–1968, Sept. 2007.

Campbell, J.P.; Shen, W.; Campbell, W.M.; Schwartz, R.; Bonastre, J.-F.; Matrouf, D., "Forensic speaker recognition," Signal Processing Magazine, IEEE , vol.26, no.2, pp.95-103, March 2009



Title of Project: Automatic Speaker Recognition - understanding possibilities and limits


Speech is a competitive biometric modality for various reasons. Particularly, speech is often the only available modality for a large set of applications in the commercial field as well as in the forensic one.

During the past years, the progress observed in the speaker area were very impressive, as shown in the NIST evaluation campaigns. The level of performance reached by the best systems encourage to embed speaker recognition inside operational, real world, applications, which is of course a correct and stimulative objective.

During the same time period, several researchers representing several scientific societies sent a "need of caution" message, showing the dangers of a direct transposition of the systems and results in the forensic field.

In order to investigate this important question, our work at the LIA, University of Avignon, aims to:

Ø      provide a SoA open-software for speaker recognition (ALIZE/SpkDet), evaluated during the NIST SRE campaigns. This system allows to experiment easily the performance and limits of speaker recognition. It also proposes an easy comparison between the different proposed approaches.

Ø      analyze more deeply the performance of automatic speaker recognition systems. There is a large number of non evaluated factors which impact the performance like speaker specificities, robustness of the modeling, linguistic and phonetic content of the messages...