Rule-Based Automatic Phonetic Transcription for the Romanian Language

In this paper, we report on an implemented rule-based tool for grapheme to phoneme conversion for the Romanian language and experiments run on two different sets of words. A 4779 words database consisting of the most frequent words in a text corpus collected from 93 books representing genuine Romanian literature, foreign literary works translated into Romanian and scientific texts was transcribed obtaining more than 95% accurate phonetic transcriptions, using a set of 102 letter-to-sound rules. Then, using the same rules, a larger set consisting of 15599 words was transcribed generating 91.46% accurate transcriptions. Although the rules weren't written by phoneticians and can be further improved, it is clear that even for mostly phonetic languages like Romanian, rule-based letter-to sound systems with manually introduced rules are limited by the human inability to spot important patterns in the pronunciation.