A Comprehensive Study on Bangla Automatic Speech Recognition Systems

This article examines the current state of speech recognition technologies in Bangla, which is one of the Low Resource Languages (LRLs) spoken by a population of almost 185 million in Bangladesh and Western India. Out of the 12,338 articles on Bangla Speech Recognition Systems, 15 have been selected based on the fine-tuning approach of deep-learning algorithms, speech corpus development, gender bias mitigation, and unique acoustic modeling. However, the datasets and evaluation metrics used in these articles vary, making it difficult to compare their performance. To address this, a comparative analysis of the selected papers is provided, summarizing the technological approaches, adapted methods, and achieved performance. Additionally, a prospective application of speech-based data collection in the healthcare domain is introduced, highlighting its potential.