The Designing And Realizing Of Multi-person Collaborative Audio And Video Annotation System

With the rapid development of computer and network communication technology,audio and video multimedia data is increasing exponentially,people have put forward higher requirements to quickly and accurately find the target audio and video,and the research of audio and video content has also become a hot spot in the field of artificial intelligence.Although artificial intelligence technology has various requirements for understanding and analyzing audio and video,the labeling requirements for audio and video data information are consistent,and the labeling of audio and video content is universal.With the gradual implementation of artificial intelligence from theory to the market,the requirements for scenarios are also increasing,and the use scenarios are becoming more and more fragmented.More and more small teams have joined the research and training of artificial intelligence models.Demand for scale annotation grows.However,the small team encountered many problems in the actual selection of audio and video data annotation tools:(1)At present,there are few open source annotation tools that support both audio and video annotation;(2)Most open source annotation tools only support user local annotation,which is not conducive to team collaboration Labeling;(3)The relatively mature and commercial labeling platforms at home and abroad have complex functions and permissions,are not friendly to team members who do not have a professional background in artificial intelligence,and are not suitable for agile labeling of small teams;(4)The team has a relatively high demand for privatized deployment.High,most commercial artificial intelligence development platforms are expensive to deploy.Therefore,having an annotation platform that supports multi-person collaboration,is lightweight,easy to use,and portable is of great significance to the research of small-team AI.In response to the above problems,this paper designs and implements a Web-based multiperson collaborative audio and video labeling system,which is specially designed to simplify the team labeling process and help small teams to flexibly and efficiently label large-scale audio and video data.The main work of the paper is as follows:(1)Design of multi-person collaborative audio and video annotation system.Based on the research on audio and video datasets,audio and video annotation tools,and multi-person collaborative annotation platforms,this paper analyzes the functional and non-functional requirements of the system,and designs the overall functional structure of the system,which mainly includes user management,data set management.(2)Implementation of multi-person collaborative audio and video annotation system.This system builds the overall architecture of the server based on the Nest.js framework and My SQL database,uses the Next.js framework and combines the Wavesurfer.js audio visualization library to complete the front-end development,and deploys the system through Docker privatization,and finally realizes the system’s control of users,data sets and tasks.The visual operation management of the workbench is convenient for the team to complete related operations such as marking,auditing and acceptance.(3)Function and performance test of multi-person collaborative audio and video annotation system.After the system development is completed,this paper deploys the system test environment to test the function and performance of the system modules.Combined with the analysis of the test results,the design and implementation of the multi-person collaborative audio and video annotation system are summarized,and further development plans are made in the future.