FABIOLE, a Speech Database for Forensic Speaker Comparison

FABIOLE, a Speech Database For Forensic Speaker Comparison Moez Ajili, Jean-Franc¸ois Bonastre, Juliette Kahn, Solange Rossato, Guillaume Bernard University of Avignon, National Metrology and Testing Laboratory , University of Grenoble Avignon, Paris, Grenoble {moez.ajili, jean-francois.bonastre}@univ-avignon.fr, {juliette.kahn,guillaume.bernard}@lne.fr, solange.rossato@imag.fr Abstract A speech database has been collected for use to highlight the importance of “speaker factor” in forensic voice comparison. FABIOLE has been created during the FABIOLE project funded by the French Research Agency (ANR) from 2013 to 2016. This corpus consists in more than 3 thousands excerpts spoken by 130 French native male speakers. The speakers are divided into two categories: 30 target speakers who everyone has 100 excerpts and 100 “impostors” who everyone has only one excerpt. The data were collected from 10 different French radio and television shows where each utterance turns with a minimum duration of 30s and has a good speech quality. The data set is mainly used for investigating speaker factor in forensic voice comparison and interpreting some unsolved issue such as the relationship between speaker characteristics and system behavior. In this paper, we present FABIOLE database. Then, preliminary experiments are performed to evaluate the effect of the “speaker factor” and the show on a voice comparison system behavior. Keywords: forensic voice comparison, intra-speaker variability, speaker recognition, FABIOLE database. 1. Introduction Speaker recognition (SR) system comparison is made pos- sible by the organization of evaluation campaigns such as those organized by the National Institute of Standards and Technology (NIST), NIST-SRE. Since the first organiza- tion in 1996, SR systems have achieved significant pro- gresses and have reached low error rate (≈ 1%). The ro- bustness of this kind of evaluation is ensured by using a very large number of voice comparison samples. The num- ber of samples per speaker as well as the characteristics of the speakers themselves, except for their native language and sex, are not taken into account in the evaluation plan. A major challenge for present SR systems is their toler- ance to speaker differences and variations in context. Dif- ferences in speaker characteristics are a major source of inter- and intra-speaker variation which should be taken into consideration. Some research have attempted to deal with some aspects of this variability (Kahn et al., 2010; Doddington et al., 1998) in which authors have shown that speaker recognition performance depends on “speaker fac- tor”. If inter-speaker variability is an important factor, intra-speaker variability is not less important (Kahn et al., 2010). It involves many factors as speaker accent or dialect, speaking style, prosody, emotion and even speaker age. Re- garding the latter, (Matveev, 2013) show a clear trend of degradation of the performance of automatic speaker recog- nition systems in a time interval of up to 4 years. Conse- quently, both inter- and intra-speaker variability should be studied in order to have a reliable assessment. These infor- mation could prove useful in some sensitive applications such as forensic voice comparison. The availability of good speech databases is crucial for “speaker factor” assessment. SR evaluation is …

Content truncated to 500 words