Setting up with the MediaLive API or CLI

The only change a function output needs on the MediaLive side is the feed ARN. There is no video description to add (that is smart crop), no caption selector (that is smart subtitles), and no output group.

{
  "InferenceSettings": {
    "FeedArn": "arn:aws:elemental-inference:us-west-2:111122223333:feed/feed-id"
  }
}

If the function reads media.transcript, map the feed's audio input to a channel audio selector so that speech recognition has audio to work with.

{
  "InferenceSettings": {
    "FeedArn": "arn:aws:elemental-inference:us-west-2:111122223333:feed/feed-id",
    "AudioFeedInputs": [
      { "FeedInput": "default-audio", "AudioSelectorName": "Audio_1" }
    ]
  }
}

If the input attachment has no audio selectors, you can omit AudioFeedInputs and MediaLive uses the input's default audio. For more information, see AWS Elemental Inference in the AWS Elemental MediaLive User Guide.