Bengali text to international phonetic alphabet (IPA) transcription: a comprehensive BYT5 based approach

Abstract

The International Phonetic Alphabet (IPA) is an alphabetic system of phonetic notation based primarily on the Latin script. The IPA is widely used by lexicographers, foreign language students and teachers, linguists, speech-language pathologists, singers, actors, language contractors, and translators. In the rapidly evolving digital era, the use of the International Phonetic Alphabet for the Bangla language is crucial to maintaining linguistic precision, aiding language learning, fostering research, preserving linguistic diversity, and facilitating communication and speech therapy. It plays a vital role in various aspects of the study of the Bangla language and related fields. In the work, it is planned to train T5-based Large Language Model (LLM) architectures on the collected dataset to convert Bangla texts into IPA notation and compare the LLM results to find the best language model. Our work extends beyond training, delving into the exploration of potential enhancement avenues. Another aspect is to investigate the impact of Bangla numerals, abbreviations, English texts, and out-of-vocabulary words. We trained and evaluated three T5-based architectures—ByT5, UMT5, and mT5—on a merged dataset of over 50,000 Bangla sentences representing both standard and dialectal speech. Through these explorations, we observe a spectrum of outcomes, where some modifications result in tangible performance improvements, while others offer unique insights for future endeavors in the fields of Natural Language Processing (NLP) and linguistic preservation.