Information is passed between tasks in the form of tracks.

Tasks usually produce multiple tracks. For example, an ingest task can produce an image track that contains frames from the ingested video, and an audio track that contains the audio packages.

In the following example, an ingest task takes the source video and produces an image track and an audio track. The image track is used as the input for an analysis task. This could be object recognition, OCR, or another type of analysis that operates on images. In this example the audio track is not used. The analysis task produces some metadata tracks which contain information about the video content. One of the tracks is used as the input for an output task.

When you configure a task, you might need to specify the track(s) to use as the input for the task:

Task type Default input tracks
Ingest Ingest tasks do not accept input tracks. Specify the source media in the process action, when you start processing.

Encoding tasks automatically use the first image and audio tracks produced by your ingest task, so you only need to specify the input for an encoding task if you want to encode different data, for example:

  • Some video sources contain more than one audio stream, to supply audio in multiple languages. You might want to encode an audio stream other than the default.
  • You might want to encode data produced by another task. For example, you might want to encode the keyframes identified during keyframe analysis, instead of all ingested frames.
Analysis Most analysis tasks automatically analyze the first image or audio track produced by your ingest task, so in most cases you do not need to specify an input track. However, some analysis operations require additional data. For example, face recognition requires the metadata produced by face detection, so when you configure face recognition you must specify an input track.
ESP You must always specify the input track(s) to use.
Transform You must always specify the input track to use.
Output Output tasks automatically use the default output tracks produced by your analysis tasks, but you can specify the input track(s) so that the output tasks use different tracks.