Meetings are important venues for documenting discussions efficiently in academia, industry and corporate settings, and the need for automatic transcription and summarization of meetings has increased. In this paper, we present SmartMeet, an end-to-end intelligent meeting minutes generation system, which exploits Automatic Speech Recognition (ASR), Optical Character Recognition (OCR) and Natural Language Processing (NLP) to automatically transcribe meetings, extract metadata and generate concise summaries. The proposed framework is evaluated on the AMI Meeting Corpus, which consists of realistic multi-speaker meeting recordings and manually annotated transcripts. We benchmark several transformer-based models on ASR, extractive and abstractive summarization tasks with standard evaluation metrics to find the most appropriate models. The generated transcripts are processed using transformer-based extractive and abstractive summarization techniques to produce structured meeting minutes. Experimental results show that Whisper achieves the best transcription performance, and our proposed extractive summarization model achieves a ROUGE1 score of 54.27, ROUGE-2 score of 49.75, ROUGE-L score of 34.17 and a BERTScore F1 of 69.48, which demonstrate an effective lexical overlap and strong semantic similarity with the reference summaries. The proposed SmartMeet framework offers an scalable solution for meeting documentation, reducing manual effort while improving the accuracy of meeting transcription and summarization.

Automatic Speech Recognition (ASR),AMI Meeting Corpus,Abstractive Summarization,Natural Language Processing,Extractive Summarization,Transformer Models,Whisper