Skip to main content

Collect a sign-language video corpus

  1. Projects → Create project. Name it on the Project Name tab.
  2. On Labeling Setup, open the Sign Language Collection group and pick a template for your elicitation style:
    • Sign this prompt (video corpus) — a written word, phrase, or sentence, signed to camera
    • Sign a picture (video corpus) — an image prompt, for signers of any written-language literacy
    • Prompted question (spontaneous signing) — a question answered in the signer's own signs
  3. Still on Labeling Setup, adjust the recorder if needed: all three cap clips at maxDuration="30"; add skeleton="hands" (or face, all) to capture per-frame keypoints next to each clip. Use Test camera / Test tracking in the preview to check framing — nothing is recorded there.
  4. On Data Import, upload one task per prompt — a CSV whose column matches the template's $variable: text for Sign this prompt and Prompted question, image (a URL or uploaded file) for Sign a picture. Save.
  5. Contributors record: camera and microphone are requested on first use, the preview shows the live frame (and skeleton), recording auto-stops at the cap, playback shows duration / resolution / frame rate, Re-record discards the take. Clips recorded offline or on a dropped connection are saved on the device and upload themselves later. On mobile the same flow runs once Allow on mobile is on.
  6. Validate in the review stream, or switch the project's QC pipeline on to get container checks plus the Listen crowd loop with a video player — see Validation; the Audio QC settings tab is there for video projects too.
  7. Export with Download ZIP with media — annotations.json plus every clip and pose sidecar in media/.

Go deeper​