In the context of the ACMus research project, we are investigating automatic techniques for annotation and segmentation of digital archives of non-western music. In particular, we are focusing on a collection of traditional Colombian music compiled in the Músicas Regionales archive at the Universidad de Antioquia in Medellín, Colombia. Of particular interest are a series of recordings of vocal expressions of indigenous cultures native to different regions of the country. These vocal expressions can either be very close to speech (almost asreciting something), can include some melodic elements, or can be closer to the concept of singing voice from a Western music perspective (exhibiting a defined melodic line). From anautomatic classification point of view, the traditional binary discrimination between speechand singing voice falls short, calling for a more general characterization of vocal expressions. For example, one may use a range[0,1] with 0 being pure speech,1 being singing voice (from a Western music perspective), and allowing everything that falls in between these bounds. In this context, interesting research questions arise: Is the degree of melodic elements in these vocal expressions informative for the region where a recording was made? Can we conclude the functional aspects of these recordings (ritual, prayer, social, playful, healing) based on this characterization? Could these categories be shared between different cultures?