Define who must be visible
A conversation recording may show the listener, a gallery grid, or a presentation slide while someone speaks. Specify whether the active speaker must be visible and whether multiple faces are acceptable. Distinguish a visible person from a verified speaking person. AVA ActiveSpeaker is a primary research example of treating speaking activity associated with visible faces as a distinct annotation task.
Specify useful framing
Describe camera distance, face size, viewpoint, occlusion, and whether hands or upper-body gestures matter. Include the effect of screen sharing and automatic camera switching. A montage of attractive frames does not establish consistent visibility over time. Ask for full sample intervals and record when required facial or body regions are absent rather than assuming the entire recording has one framing condition.
Inspect audiovisual correspondence
Check the delivered sound against the visible activity and note voiceovers, dubbed segments, music, and off-screen speakers. If multiple audio tracks are offered, identify the intended track and its relationship to the visual scene. Keep speaker labels, face tracks, and timestamps linked to the exact export. Describe uncertainty instead of assigning identities unsupported by the source or review process.
Review intended use specifically
Analysis of talking-person footage and generation involving likeness or voice raise different project requirements. Explain the intended training use and request relevant source and participant documentation for review. A recording being available for viewing does not establish suitability for every use. Fiund discusses availability, samples, documentation, and commercial scope for each proposed project rather than offering a blanket permission claim.
Review checklist
- Require visible presence or active speaking explicitly.
- Specify face and body framing over time.
- Check the exact soundtrack against visible activity.
- Describe intended use and required source documentation.