The wide range of potential commercial applications for a com puter system capable of automatically converting text to speech (TTS) has stimulated decades of research. One of the currently most successful approaches to synthesising speech, concatenative TTS synthesis, combines prerecorded speech units to build full utterances. However, th e prosody of the stored units is often not consistent with that of the target utterance and m ust be altered. Furthermore, several types of mismatch can occur at unit boundaries and must be smoothed. Thus, pitch and time-scale modification techniques as well as smoothing algorithms play a critical role in all concatenative-based systems. This thesis presents the developm ent of a concatenative TTS system based on a harm onic model and incorporating new pitch and time-scaling as well as smoothing algorithms. Experim ent has shown our system capable of both very high quality prosodic modification and synthesis. Results com pare very favourably with those of existing state-of-the-art systems.