Persian Automatic Speech Recognition by the use of Whisper Model

Automatic speech recognition (ASR) systems play a pivotal role in modern communication technology by enabling computers to convert human speech into written form. The Whisper model has demonstrated remarkable performance in ASR, leveraging advanced deep learning techniques and training on extensive multilingual datasets. However, applying the Whisper model to Persian ASR presents unique challenges due to the Persian language’s complex phonological features and non-Latin Persian script. This paper aims to maximize the performance of the Whisper model in Persian ASR by fine-tuning the model and adjusting parameters. Experimental results demonstrate that the proposed method outperforms other approaches significantly. To advance research in this field, we make all the source code of the proposed method available at GitHub: https://github.com/mohammadhl28/Persian_ASR.