Merging segmental, rhythic and fudamental frequency features for Automatic Language Identification

This paper deals with an approach to Automatic Language Identification based on rhythmic and fundamental frequency modeling. Experiments are performed on read speech for 5 European languages. They show that rhythm can be automatically extracted and is relevant in language identification: using cross-validation, 79% of correct identification is reached with 21 s. utterances The fundamental frequency modeling, tested in the same conditions (cross-validation), produces 50% of correct identification for the 21 s. utterances. The Vowel System Modeling gives an identification rate of 70% for the 21 s. utterances. Last, merging the three models slightly improves the identification rate.