The present study discusses recent efforts to improve the segmental quality of the Swedish text-to-speech system developed at KTH. The emphasis is put on the study of consonant sequences as found in running speech. Rules working inside words and across word boundaries are exemplified.