Abstract: A multi-language text-to-speech system developed by Carlson and Granstrom was used to do text-to-speech conversion for Chinese. Both phonological rules and scoustic parameters are refined. The light tone and tone modification rules and er-coloured final rules are introduced. It is shown that to shorten duration is the most important feature for light tone syllables, and the principle of that the main vowel in the final should be pronounced with retroflex vowel /er/ simultaneously can be used for synthesizing er-coloured finals.