This work represents the first known attempt to develop a text-to-speech synthesiser for Welsh. A list of pseudo-Welsh nonsense words was generated, allowing for certain difficulties particular to Welsh. Diphones were derived semi-manually from the recorded nonsense words. The first known phonemic lexicon for Welsh was derived from an electronic corpus. Letter-to-sound rules for Welsh were written, differentiating between the vocalic and consonantal realisations of two graphemes, and assigning lexical stress. An existing English text-to-speech synthesis system was adapted for Welsh. Some simple duration and F0 rules were written that gave pleasing results with the minimum of rules. The resulting system can be used for Welsh or for English spoken with a recognisable Welsh accent.