This paper describes the microphonemic speech synthesis method and its implementation in the form of a text-to-speech synthesizer. The synthesis is based on combining waveforms that are collected from real speech, thus retaining the most part of the individual features of the speaker. The rule-based text-to-speech synthesis is based on object-oriented representations of the hierarchical structure of the units in speech. Syllable classes play an important role as a unit for the prosodic structure.