Most of business around the globe hold most of their meetings online. Videoconferencing is used for meeting with customers and partners as well as for internal meetings, trainings and conferences. Many of them record their meetings for future reference. Companies have accummulated hundreds of meeting recordings and store terrabytes of videodata.

Internal users access videoconference recordings for various reasons: to recall details and numbers of somebodies speach, to clarify someones commitment, to recall decisions and conclusions on specific topic discussed during the meeting. Browsing through videoconferece recordings is very time-consuming as user has to look and listen to most of meeting to avoid missing target information.

Trying to find information on specific topic without knowing exact day and time of meeting appers to be impossible as well as finding meetings with specific participant. Therefore, storing meeting recordings creates much less value then it could.

Vision Systems has created an intelligent video analytics solution based on Computer Vision and Natural Language Processing technology. The system analyzes the videoconference recordings and supplements video with metadata to convert video storage into structured and searchable knowledgebase. Machine learning algorithms use face, articulation and speech recognition technologies to convert audio to text and match particular phrases to speakers. We also built a user portal allowing users to filter recordings by participants, seerch by keywords pronounced during the meeting and navigate directly to the targeted moment of recording from the search result page.

Navigation within one videconference recording allowed users to see timestamps of selected speaker activity and jump directly to required fragment.

A simple sorage of videofiles has turned into powerfull knowledgebase allowing users to find information they need easyer and faster. This fact leads to making better decisions, faster learning process and making less mistakes. Commercial efficiency on data storage increased many times.

Saved videoconferences are processed by few Machine Learning algorithms: #NLP transcribes speach to text, next algorithm performs #FaceRecognition, other identifies speakers activity by lip movement to bind phrases to appropriate speakers. Collected information is structured to metadata related to video and stored in database.

User self-service portal allows users to sort and filter videos by participants, topics, video name, date and length. Portal internal search engine helps users find specific moments inside videos and navigate directly to required video fragments by clicking item in search results.

System can evolve by adding new features: gesture recognition for bookmarking or other purposes, specific object recognition or detection, etc. There are many directions for systems growth. For example, we can add functionality of automatic e-mailing meeting minutes with commitments and meeting conclusion to meeting participants.