A novel kind of neural network architecture is called a transformer. The HuggingFace transformer models offer a user-friendly way to deploy few top-performing Natural Language Processing (NLP) models. In a number of NLP applications, including text categorization, text generation, text summarization, and question answering, transformers models are accepted as the state of the art. The fact that it has improved the efficiency and accuracy of jobs like natural language processing is what's drawing attention. One such transformer model that trains the pre-trained models is the Bidirectional and Auto-Regressive Transformers (BART) model, which borrows elements from various transformer models. Sequence-to-sequence denoising was proposed in the BART pre-training for the paper on comprehension, translation, and natural language generation. Pre-trained weights and adjusted weights for text summarization, question answering, conditional text generation, sequence classification, and mask filling can be utilized with the BART model. In this paper, first speech is taken as input to be converted as text brilliantly then many features of the BART implemented successfully implemented with > 99% accuracy. Python language is used with BART transformer for sec2sec, paraphrasing, tokenizing with pipeline utilities.