In the world of artificial intelligence, it has become essential to understand the natural language of human beings from anywhere in the world. Speech recognition enables machines to understand human communication better. Therefore, many speech recognition system has been developed and remarkable progress is noticed in the last few decades. Language analysis and readability are essential for native speakers to connect with the digital world. This work will enable global users to read Assamese transcription effectively by facilitating its pronunciation with the help of International Phonetic Alphabet (IPA) symbols without learning the Assamese script. The motive of the work is to develop an automatic system that could be able to translate speech in Assamese language into its correct corresponding IPA representation. A CNN-based model wav2vec2-XLS-R-300 M is transformed and fine-tuned through a dataset containing Assamese speech and corresponding transcriptions. To confirm a vigorous efficiency of the model, the dataset is divided into three disjoint sets of training, testing, and validation. Our model achieved 86% performance accuracy at character level and 14% character error rate (CER). However, a high word error rate (WER) of 96% and F1 score of 0.03 highlight the complexity in word level and underscore the need for improvisation in handling word boundaries and contextual nuances in Assamese speech. Insights gained from this project lay a foundation for future scope in enhancing inclusive automated speech transcription systems for underrepresented languages, contributing to linguistic research and language preservation.