A LLM-Powered Approach for Generating Meeting Minutes from Audio Conversations

Accurate documentation of meetings is essential for effective communication, decision-making, and accountability across various settings, including corporate environments, academic discussions, research collaborations, project reviews, and multi-stakeholder forums. Manual note-taking is often error-prone and time-consuming, highlighting the need for automated solutions. This paper proposes an LLM-based approach for generating structured Minutes of Meeting (MoM) from audio recordings. The method begins with audio transcription using the Vosk model, followed by grammatical and punctuation recovery to enhance text readability. The preprocessing step cleans and segments the transcribed text. Named Entity Recognition (NER) is applied to identify relevant entities such as names, dates, and organizations. The method then segments the conversation into topical sections aligned with the meeting agenda. A Large Language Model (LLM), named OpenAI, has been used to produce abstractive summaries. An action item extraction module identifies tasks, responsible individuals, and deadlines. A speaker analysis component highlights principal participants in decision-making processes. The system organizes the final output into a structured format, such as JSON or PDF, for easy access and distribution. This end-to-end approach enhances the clarity, accessibility, and utility of meeting documentation, reducing manual effort and improving organizational productivity.