Deep Transfer Learning based Speech Recognition for Low Resource Hindi Language

Automatic speech recognition (ASR) stands out as a pivotal technique, steadily gaining prominence by facilitating seamless communication between humans and technology. Due to a shortage of substantial speech datasets for model training, ASR systems for most languages lack acceptable performance, despite achieving human-like performance for a limited number of languages. In this study, we leverage the Hugging Face ecosystem to fine-tune state-of-the-art Whisper Speech architecture to showcase the efficacy of transfer learning technique in ASR tasks, yielding momentous enhancements in accuracy and performance compared to conventional training methods. Our findings underscore the potential of transfer learning to advance ASR technologies and provide insights for future research endeavors in this dynamic domain.