Expressive Performance in the Human Tenor Voice

This paper presents preliminary results on expressive performance in the human tenor voice. This work investigates how professional opera singers manipulate sound properties such as timing, amplitude, and pitch in order to produce expressive performances. We also consider the contribution of features of prosody in the artistic delivery of an operatic aria. Our approach is based on applying machine learning to extract patterns of expressive singing from performances by Josep Carreras. This is a step towards recognizing performers by their singing style, capturing some of the aspects which make two performances of the same piece sound different, and understanding whether there exists a correlation between the occurrences correctly covered by a pattern and specific emotional attributes.