Language Identification in Vocal Music

Language identification is an important field in spoken language processing. The identification of the language sung or spoken in music, however, has attracted only minor attention so far. This, however, is an important task when it comes to categorizing, classifying and labelling of music data. In this paper, we review our efforts of transferring wellestablished techniques from spoken language identification to the area of language identification in music. We present results of distinguishing German and English sung modern music and propose and evaluate techniques designed for improving the classification performance. These techniques involve limiting the classification on song segments that appear to have vocals and on frames that are not distorted by heavy beat onsets.