The application of several deep-learning approaches has led to remarkable improvements in automatic speech recognition (ASR). In this paper, the authors transcribed recordings of OOD-Speech. For Bengali automatic speech recognition (ASR) with respect to the training set using the Whisper model and the Multilingual Automatic Speech Recognition approach, the first out-of-distribution (OOD) benchmarking dataset is available. The authors implemented an English text normalizer to standardize the transcription and computed the Word Error Rate (WER): 4.27%; which is a very low error rate and indicates highly accurate Bengali speech recognition. This work is part of a code competition of Bengali.AI Speech Recognition on Kaggle to recognize Bengali speech from out-of-distribution audio recordings.