A set of letter-to-sound rules was written, in the form of context-sensitive rewrite rules. A publicly-available program was used that converted this style of rule to a C program. Three sets of rules were written, reflecting three separate passes through each input word. The first set added epenthetic vowels, while the second set located the vowels and lexical stress. The third set carried out grapheme-to-phoneme conversion. The rules were evaluated over text from a magazine article, and found to have a 96% success rate.
Keywords: Text-to-speech synthesis, Welsh, letter-to-sound rules.