Advances in Automatic Meeting Minute Generation: A Survey

Abstract

We faced the largest crisis of the twenty-first century at the start of 2020: the COVID-19 pandemic. In the midst of the turmoil, the generation ultimately found a method to get the job done by using automation in many aspects of life. Following the epidemic, we saw an 87% increase in video conferencing technologies for daily communications. Almost everything, from online gatherings to college lectures to business meetings, was housed on the internet, which, because it was virtual, increased the odds of ineffective interactions. In reality, statistics collected from employees across all domains reveal that people frequently miss essential points since taking minutes of meetings is a time-consuming, distracting, and extremely dull chore, and that over 37 billion dollars is squandered on ineffective meetings. Keeping track of significant decisions and agreements that were reached at a meeting requires the use of meeting minutes. The issues addressed and the choices made must be recorded in order to be reviewed at the start of the following meeting and for future reference. Many businesses retain salaried personnel to take minutes of meetings, using up valuable time and resources. We provide a method to enable staff members to have productive conversations that will increase a company's productivity by making greater use of the tools and technical improvements that are now accessible. Our approach extracts crucial information from significant debates using Deep Learning methods. The suggestion is for an automated method to record minutes and transcripts of a meeting with the benefit of speaker identification. The model we suggest will be able to recognise the speaker using Mel Frequency Cepstral Coefficient (MFCC)[12], convert an audio file into plain text using Deep Neural Networks (DNN), and summarise the meeting transcript into condensed minutes with the aid of Transformers.