Introduction

News segmentation analyzes news broadcasts, identifies the times at which stories start and finish, and extracts the key concepts from each story.

News segmentation classifies the ingested video into segments that can be:

  • A story - a video segment that contains a consistent set of concepts.
  • A short story - a video segment that contains a consistent set of concepts, but the total length is too short to be considered a complete story. For example, a news presenter might briefly describe the news headlines at the end of the hour, so each topic is likely to be covered for only 20 or 30 seconds.

  • No topic - a video segment that does not contain a consistent set of concepts.

The news segmentation task is intended to be used on speech that has been transcribed from the audio track of a news broadcast, and segmented into sentences by a text segmentation task.

News segmentation enables Media Server to output documents that correspond to stories in a news broadcast. When you configure Media Server to output documents, use bounded event mode and use the news segmentation result track as the event track. For more information about configuring Media Server to output documents, see Output Data.