Regional-Speech: A Bengali Speech Recognition Dataset for Benchmarking Models Under Dialect Variation in Puran Dhaka.

Abstract

Regional dialects of the Bengali language, which is spoken throughout South Asia and among the Bengali diaspora, are influenced by historical, cultural, and geographic factors. Eastern Bengali, Manbhumi, Rangpuri, Varendri, and Rarhi are the five main dialects of Bengali based on phonology and pronunciation. There are additional differences in vocabulary, pronunciation, syntax, and morphology within Bangladesh. The distinctive characteristics of dialects from areas such as Dhaka, Chittagong, Sylhet, Rangpur, Rajshahi, Noakhali, and Barishal set them apart from both standard Bengali and from each other. Notwithstanding this linguistic diversity, there is still a dearth of resources and research devoted to comprehending regional Bengali dialects and incorporating them into natural language processing (NLP) systems. By examining these dialects using thorough, datadriven linguistic analyses, such as phonetic and morphological studies, this study seeks to close this gap. We also evaluate the viability of creating computer models customized for these dialects, such as Automatic Speech Recognition (ASR) systems. Applications like virtual voice assistants and other Bengali language tools might be made possible by such models. Our research aims to promote inclusivity and effective communication while advancing knowledge and supporting the preservation of regional Bengali dialects (Puran Dhaka “Dhakaiya”). In order to ensure the Bengali language's continued relevance in contemporary computational applications, this research helps to develop language technologies that respect the language's cultural and linguistic heritage by attending to the linguistic needs of Bengali-speaking communities.