Towards Automatic Identification Of Singing Language In Popular Music Recordings

The automatic analysis of singing from music is an important and challenging issue within the research target of content-based retrieval of music information. As part of this research target, this study presents a first attempt to automatically identify the language sung in a music recording. It is assumed that each language has its own set of constraints that specify which of the basic linguistic events present in a singing process are allowed to follow another. The acoustic structure of individual languages may, thus, be characterized by statistically modeling those constraints. To this end, the proposed method employs vector clustering to convert a singing signal from its spectrum-based feature representation into a sequence of smaller basic phonological units. The dynamic characteristics of the sequence are then analyzed by using bigram language models. Since the vector clustering is performed in an unsupervised manner, the resulting system does not use sophisticated linguistic knowledge and, thus, is easily portable to new language sets. In addition, to eliminate the interference of background music, we leverage the statistical estimation of a piece’s music background so that the vector clustering is relevant to the solo singing voices in the accompanied signals.