Proceedings of the Annual Congress of the Brazilian Computing Society / VIII Brazilian Symposium on Computer Music, Fortaleza, Brazil. 2001 Computer-Aided Song Design: Prosody as Scaffolding Eduardo Reck Miranda SONY Computer Science Laboratory Paris 6 Rue Amyot – 75005 Paris – France miranda@csl.sony.fr - http://www.csl.sony.fr Abstract Composers often face the task of composing melodies for given texts such as religious hymns, poetry or the libretto of an opera. A plausible point of departure for writing melodies for a text is to study the prosody of the text as spoken either naturally or dramatically. In this paper we introduce PROSE: a system for aiding such study. The system extracts the prosody of a spoken signal and (re)synthesises it at various resolutions. The main advantages of using PROSE over simply listening to the spoken signals are that composers can focus on prosodic auditory information detached from the meaning of the text and can assess this information at various resolutions. Also, the analysis data can be plotted for visual assessment and/or mapped onto musical parameters. 1. Introduction Composers often face the task of composing melodies for given texts such as religious hymns, poetry or the libretto of an opera. Different composers have approached this problem in a variety of ways, ranging from sheer inspiration and intuition to formal approaches. Formal approaches to song writing in Western European music are as old as music theory itself. One of the finest examples of early formal approaches to song writing appeared in the eleventh century, when Guido d’Arezzo proposed a lookup chart for assigning pitch to the syllables of religious hymns. He also invented the musical stave for systematic notation of music and established the medieval scales known as the Church modes (Sadie 1981). Although computer music systems are normally orientated towards formal approaches to composition (Miranda 2001), in this paper we propose the use of the computer as a powerful aid to inspiration, rather than as a processor of formalisms for generative song writing. Writers and poets often assess their work by reading them aloud: in these cases a text is finally right when its intonation sounds right. What does “sounds right” mean here? What makes the placement of a word “x” in a sentence sounds better than a word “y” of identical lexical meaning? There are no simple answers to these questions. Indeed, these are hot topics in current Linguistic research and surveys of the intonation systems of various languages using a common methodology have already begun to appear (Hirst and DiCristo, 1998). Still, very little is known about the overall intonation properties of different languages. What we do know, however, is that prosody seems to hold an important key to unlock these questions. Composers, on the other hand, do not usually write the texts for their songs, but work with given texts. Some texts are perhaps easier than others to work with because their intonation properties were better explored by the writer. A plausible point of departure for writing melodies for a text is therefore to study the prosody of the text as spoken either naturally or dramatically. What do we mean by “prosody”? How can the computer aid composers to study the prosody of a text? The literature on phonetics often defines prosody primarily in terms of the