End-to-end Automated Speech Recognition (ASR) has recently benefited from advancements in multimodal language models, which combine audio and visual data to improve transcription accuracy and context understanding. Neural networks have become increasingly popular for the task of language modeling. A language model (LM) is a mathematical representation of the likelihood of a word sequence that offers a solution to word prediction in different types of information systems. Learning the long-span dynamics of a continuous-space word sequence is a powerful task for a Recurrent Neural Network (RNN). On the other hand, with a high-dimensional hidden layer and a huge dictionary size, training the RNN-LM becomes an ill-posed task. In contrast to traditional unidirectional networks, which rely solely on historical data to provide predictions, bidirectional networks can take into account both current and future inputs. In this research, we suggest using bidirectional Recurrent Neural Networks (BRNN) for language modelling in speech recognition. We examine bidirectional models on an Indian English Broadcast News transcription assignment and talk about problems that come up when using them for speech. With the use of testing data from 50 epochs, the audio speech of proposed dataset was accurately recognized with a 97.74% accuracy rate and a precision rate of 95.23%. Hence, multimodal BRNN-based ASR framework offers a robust approach to end-to-end speech recognition, providing a foundation for further exploration into multimodal enhancements.