Can Automatically Extracted Rhythmic Units Discriminate among Languages?

This paper deals with rhythmic modeling and its application to language identification. Beside phonetics and phonotactics, rhythm is actually one of the most promising features to be considered for language identification, but significant problems are unresolved for its modeling. In this paper, an algorithm dedicated to rhythmic segmentation is described. Experiments are performed on read speech for 5 European languages. Several algorithms are compared. They show that salient features may be automatically extracted and efficiently modeled from the raw signal: a linear discriminant analysis of the extracted features results in a 80 % percent of correct language identification for the 5 languages, using 20 s duration utterances. Additional experiments reveal that the automatic rhythmic units convey also speaker specific features.