Developing a Thai Name Pronunciation Dictionary from Road Signs and Naming websites

A pronunciation dictionary is crucial for developing grapheme-to-phoneme conversion systems. Traditionally, most databases rely on linguists to transcribe Thai words into phoneme sequences. However, conventional dictionaries have significant drawbacks: they are costly, time-consuming, and often not up-to-date. This paper addresses these issues by focusing on creating an up-to-date pronunciation dictionary using data collected from naming websites and road signposts in Thailand, ensuring the correctness of the dictionary. This paper introduces a procedure for data preparation and model development. This includes the collection and aggregation of pronunciation data, the training of a transformer grapheme-to-phoneme model. The total number of Thai names includes approximately 19,000 pairs of unique words and phoneme transcriptions. This Thai name pronunciation dictionary is designed for use in grapheme-to-phoneme conversion methods in both speech synthesis and speech recognition. Results show that using this pronunciation dictionary with a transformer model achieves approximately 97% accuracy for exact matches and 99% accuracy when ignoring tones.