Fully automatic text-to-speech systems must accept as input any texts in whatever form they might be stored on a computer. As such, the role of punctuation characters in marking sentences, phrases and other textual constructs has to be exploited to produce natural sounding synthetic speech. Some characters not in the alpha-numeric set can, however, act both as text and as punctuation in different situations. A pre-processing module has therefore been implemented which is sensitive to these different roles and attempts to use them in preparing texts for text-to-speech conversion.