Development Of A Kiswahili Text To Speech System

Abstract

This paper discusses how a concatenative Kiswahili Text to Speech System (TTS) was developed based on the Festival Unit Selection Speech Synthesiser. It explains how important Kiswahili linguistic features such as phones, stress and intonation were modelled as inputs to the Festival engine. It also discusses the design, recording and segmentation of the speech database, beginning with text corpus collection and transcription. The choice of the speaker, which is crucial to realising a good TTS is discussed and also how the system was tested.