A Unified Pipeline for Automated Meeting Documentation Using Speech-to-Text, Abstractive Summarization, and Sentiment Analysis
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Automated meeting documentation requires jointly addressing transcription noise, long-context summarization, and affect interpretation in multi-speaker conversations. This paper investigates an end-to-end pipeline that combines speech-to-text, contextual abstractive summarization, and segment-level sentiment labeling to generate actionable meeting minutes. The system uses Whisper for transcription, a transformer encoder-decoder summarizer (T5) with a long-transcript chunking strategy, and a hybrid sentiment module based on VADER and RoBERTa to improve robustness under conversational language. Performance is evaluated using Word Error Rate (WER) for transcription, ROUGE for summarization quality, and F1-score for sentiment classification, together with qualitative feedback from agile project managers. Results show that the pipeline produces readable transcripts and coherent summaries for multi-party meetings, while sentiment labeling provides useful polarity cues that help reviewers identify tense segments and interpret discussion tone. Expert feedback indicates the outputs support faster post-meeting review and clearer follow-up alignment. The work contributes a deployable unified architecture that operationalizes summarization and sentiment analysis within a single meeting-minutes workflow and highlights concrete constraints-overlapping speech, domain jargon, and subtle affect-that guide future improvements in domain adaptation and emotion-aware meeting analytics.
Fields of Study & Subjects
Speech Recognition and Synthesis
Sentiment Analysis and Opinion Mining
Topic Modeling
Publication details
- Journal / source
- International Conference on Advanced Research in Computing (ICARC)
- Pages
- 1–6
- Publication date
- 18 February 2026
- Language
- EN