Automatic Running Notes Generation from Audio Lecture using NLP for Comprehensive Learning

Taking notes is vital for students as it provide a condensed what they learn during classes sessions. Good notetaking helps students understand better and remember more. Notes help students engage more deeply with the material. However, note-taking is difficult because it requires listening, understanding, and writing simultaneousely, especially when lectures are fast-paced. Students also need to learn how to decide what information is important to note down. Automatic note generation solves these problems by using technology to make note-taking easier. It means students don’t have to write everything down as they listen and lets them focus more on understanding the lecture. Thus, the proposed work presents a comprehensive solution for automatic notes generation from audio lectures, facilitating access to summarized content in both English and Telugu languages. Leveraging advanced techniques including Google’s Web Speech API employing a combination of Hidden Markov Models (HMMs) and Deep Neural Networks (DNNs), system accurately transcribes spoken content with a Word Error Rate (WER) of 8% and a Word Accuracy Rate (WAR) of ${9 2 \%}$. Subsequently, employing the TextRank algorithm, the system extracts key information and generates structured notes with precision of 0.82, recall of 0.9649, and an F1 score of 0.886. Furthermore, employing a neural machine translation (NMT) approach, the system achieves a BLEU score of 0.5789, enabling seamless translation of the generated notes into Telugu. The system provides users with the flexibility to download both English and Telugu notes in Microsoft Word format, thus offering versatile tool for accessible knowledge dissemination from audio lectures.