Multi-Person speech Curator for Minutes of Meetings refers to a system where we can generate the summary of a meeting from the audio recording of the meeting using automatic speech recognition (ASR) and NLP algorithms...
详细信息
ISBN:
(纸本)9798400708985
Multi-Person speech Curator for Minutes of Meetings refers to a system where we can generate the summary of a meeting from the audio recording of the meeting using automatic speech recognition (ASR) and NLP algorithms. The system can identify and differentiate speakers and generate transcript with time stamp. ASR is used to transcribe the audio of the meeting and then NLP algorithm used to for speaker diarization and then extract the summary from the transcript. In this paper we use Whisper API from OpenAI for speech to text conversion. The model facebook/bart-large-cnn is used to generate the summary from the transcript. The model is fine-tuned with Samsum dataset that contains about 16k messenger-like conversations. mBART model is used for summary translation. The objective of the paper is to create minutes of meeting which has concise information about the complete meeting discussions, decisions, and actions. The system has the potential to significantly save the time and resources as it eliminates manual transcript and captures all important points of the meeting.
暂无评论