The KTH rule system for singing synthesis

This article contains a description of rules controlling the singing synthesis at the Department of Speech Communication and Music Acoustics at the Royal Institute of Technology (KTH) in Stockholm. In our research, synthesis of singing has played an important part for a long period of time. The rules controlling the singing synthesiser MUSSE DIG are implemented in a programming environment originally developed for a text-to-speech system. There are context dependent rules for pronunciation of vowels and consonants, as well as rules for musical performance. The latter rules create crescendos, tempo and vibrato changes etc., depending on the musical context as defined by a score Jile. The rules were developed using an analysis-by-synthesis strategy, i.e., vocal performances are synthesised, the result is analysed and then the rules, which control the synthesis, are accordingly improved. In this article, musical rules, general rules for consonants and vowels, and for some special singing techniques are described.