This paper describes a modular approach to rule-based speech synthesis that we have been using for English dialects and have begun to extend to multiple languages. In the modular approach, a single program, divided into language-universal, language-specific/dialect-universal, and dialect-specific rule modules, is used to synthesize a number of dialects or languages. The approach is based on a new "multi-stream" phonetic model and is implemented in Delta, a fourth-generation programming language designed for developing synthesis rules based on such models. Besides the theoretical significance of such an approach, it results in more cost-effective development and implementation of synthesis rule sets than possible by writing separate programs or by modularizing within general-purpose languages such as C.