Design And Implementation Of Intelligent Meeting Minutes System

With the development of artificial intelligence technology,more and more intelligent application developped by using algorithms and models of machine learning and deep learning.The field of artificial intelligence includes robotics,speech recognition,image recognition,natural language processing,and expert systems.In our daily life,speech recognition and natural language processing are the most widely used,but these algorithms are relatively simple to use,and there is no application using two or more artificial intelligence technologies.Automatic text summarization technology is one of the most widely used techniques in natural language processing.This technology automatically summarizes and generates main content without changing the original intention of large-scale documents.In the era of massive information,the use of this technology to express the main content of information in a short text helps to alleviate the problem of information overload.Sentiment analysis is a computational study of the opinions,emotions,and attitudes of people's problems,events,or topics.Sentiment analysis technology allows machines to perceive emotions through speech and text,and in the specific context,uses the emotions of speech and text to mine potential user information and business information.This paper implements the intelligent voice conference minutes system based on speech recognition technology,text summarization technology and sentiment analysis technology in deep learning.First,the system voice processing module is composed of speaker classification module and speech recognition module.The voice processing module is used to convert the meeting audio into a meeting content.The speaker classification module uses the speaker multi-audio stream timeline strategy to classify and mark the speakers in the conference speech,and mainly studies and implements the speech endpoint detection technology.The speech recognition module mainly studies the existing excellent algorithms for speech recognition,especially the speech recognition algorithm with attention mechanism,and compares the accuracy and latency of the speech recognition model.Second,the system text module is composed of a text summarization module and sentiment analysis module.The text module takes the meeting content output by the voice module as input,extracts the summary of the meeting summary,and performs sentiment analysis on the meeting summary,and analyzes the possibility of the participants after the conference.The text abstraction module studies the text word vectorization technology and abstract extraction and generation models,including: abstractive text summarization technology and extractive text summarization,and implement the use of the latest ELMo word vector technology and attention-based abstract extraction model.Text summarization model.The sentiment analysis module studied the audio sentiment feature extraction method,the twoway LSTM technology in the emotional analysis model construction,and built the speech sentiment analysis model using the Tensor Flow deep learning framework.Finally,the system uses Python to build Web API interface,and all the modules are combined into a meeting minutes system by means of an engine.The corresponding engineering tests are carried out for different functional modules,and the system testing tools are used to test and analyze the performance of the system to verify the system and meet the requirements.