Audio and video segment labelling

When to use it

• Speaker turns and speech versus silence
• Sound events
• Notable moments in video
• Segments to transcribe

What your file looks like

Each row links one audio or video file. Short clips work best, and the segment labels come from the form. A file like this is enough:

id,clip
1,https://example.com/audio/001.wav
2,https://example.com/audio/002.wav

What reviewers see

An audio or video player, label buttons, Start here and End here buttons that capture the playhead, and a list of segments with editable times.

What a reviewer submits

A segment is a start and an end in seconds with a label, to a tenth of a second.

{
  "answer": {
    "segments": [
      {
        "label": "speech",
        "start": 1.2,
        "end": 3.4
      }
    ]
  }
}

What you get back

Segments are the same when their labels match and their overlap in time (intersection over union) is at least 0.5. The exported start and end are averages of the agreeing group.

{
  "n": 3,
  "value": [
    {
      "label": "speech",
      "start": 1.2,
      "end": 3.4
    }
  ],
  "agreement": null,
  "tie": false
}

Quality checks

• Majority filtering keeps only stretches most reviewers marked.
• Tell reviewers how to treat short pauses: split the segment or keep it whole.
• Use short clips. Marking quality drops after a minute or two of audio.

Tips

• A video column also works: the screen shows a video player instead.
• Boundary precision depends on playback. Ask for tenth-of-a-second accuracy, not more.
• Pair with a free-text question when you need a transcript for each segment.

Limits

Up to 100 segments per clip and 20 labels. Times are in seconds; video frames are not addressed individually.