This talk will look at singing and vocal production across cultures, and a number of cross-disciplinary approaches to investigate it. We’ll examine various options to formalise the language about vocal production in order to operate with it algorithmically. In particular, vocal source aerodynamics, laryngeal vibratory mechanisms and vocal tract physiology will be reviewed. We’ll discuss how new annotated datasets can be generated, under what conditions the annotations will be reliable. In relation to expert annotations important confounding issues and biases will be explored. On the example of Cantometrics – a complex comparative task from ethnomusicology – we’ll outline a possible MIR contribution, its advantages and limitations. We’ll be looking at ways to revise the experiment with contemporary MIR approaches and to scale it up to millions of recordings.