Extraction automatique de paramètres prosodiques pour l'identification automatique des langues

International audience; The aim of this study is to propose a new approach to Automatic Language Identi - cation: it is based on rhythmic modelling and fundamental frequency modelling and does not require any hand labelled data. First we need to investigate how prosodic or rhythmic information can be taken into account for Automatic Language Identi cation. A new automatically extracted unit, the pseudo syllable, is introduced. Rhythmic and intonative features are then automatically extracted from this unit. Elementary decision modules are de ned with gaussian mixture models. These prosodic modellings are combined with a more classical approach, a vocalic system acoustic modelling. Experiments are conducted on the ve European languages of the MULTEXT corpus: English, French, German, Italian and Spanish. The relevance of the rhythmic parameters and the ef ciency of each system (rhythmic model, fundamental frequency model and vowel system model) are evaluated. The in uence of these approaches on the performances of automatic language identi cation system is addressed. We obtain 91 % of correct identi cation with 21 s. utterances using all the information sources.