A COMPREHENSIVE REVIEW OF AUTOMATIC SPEECH RECOGNITION FOR ODIA LANGUAGE: CHALLENGES, TECHNIQUES, AND FUTURE DIRECTIONS

This paper highlights an extensive review of ASR for the low-resource language Odia by focusing on methodologies, datasets, metrics, applications, and future scopes. In terms of evolution, the current paper examines the progression of technology from conventional approaches like Hidden Markov Models to recent innovations, such as CNN, RNN, and attention-based systems. The challenges that have been discussed extensively within this review paper include insufficient data, dialectic issues, lack of benchmarking resources, and sensitivity to noises. The ongoing research has been primarily focused on isolated speech, although continuous speech has not received sufficient attention. Transfer learning and multilingual techniques may offer some potential in addressing existing problems.