Voice-Driven Live Notes: A Real-Time Tool for Teaching-Learning Productivity

Real-time voice-driven productivity tools have gained significant attention for their ability to transcribe and organize spoken inputs, offering immense potential for educators in managing classroom discussions and instructional materials. However, existing systems often struggle in noisy environments, with limited contextual understanding and inaccurate note categorization, reducing their effectiveness in dynamic educational settings. To overcome these challenges, this paper propose Voice-Driven Live Notes, A Real-Time Organizer for Enhanced Productivity, a framework leveraging advanced speech-to-text and machine learning technologies to achieve superior accuracy, contextual awareness, and usability. The system is built on robust datasets like LibriSpeech ASR Corpus and Common Voice, ensuring adaptability to real-world educational scenarios through diverse and noisy speech samples. It integrates Conformer and Wav2Vec <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{2. 0}$</tex> for accurate transcription in noisy environments. GPT-4 and Claude 2 provide human-like comprehension and summarization, simplifying key point extraction for teachers. Additionally, HDBSCAN and DBSCAN clustering algorithms enable noisetolerant grouping of related content for better organization. Experimental evaluations demonstrate up to 95 % transcription accuracy, surpassing existing systems. Clustering efficiency improved by 12 %, with overall productivity enhanced by 15 %. These findings establish new benchmarks in transcription precision and note organization, addressing educators' specific needs in modern classrooms.