Inputs
The face is the minimum. What the microphone, a body view and stimulation reports add, and what happens when an input is missing.
What each input unlocks
| Inputs | Imminence | Subject read | Onset and end (women) | Peak (men) | End (men) |
|---|---|---|---|---|---|
| Face | Yes | Yes | No | No | No |
| Face and microphone | Yes | Yes | Yes | Yes | Yes |
| Face, microphone and body view | Yes | Yes | Yes, corroborated | Yes, corroborated | Yes |
Declare what the device will publish when you create the session (inputs.audio, inputs.body); the session's signals then lists the events it can produce.
Face (required)
The front camera, with the person's face in view and only that person in frame.
- H.264 or H.265 video. Frames are scaled to a long edge of 1280 pixels; 30 frames a second is ideal.
- Even, frontal light. The analysis corrects for a turned or tilted head, but not for a face in shadow or out of frame.
- The first minute of visible face is the person's resting baseline. Imminence starts once it is in (
baseline_pendinguntil then); the subject read is made after the first 30 seconds.
Microphone (recommended)
Publish the microphone with the camera. The onset, peak and end detectors read the voice as well as the face and do not run without it. Sound is measured inside the enclave (classes of vocalization, level, pitch); no audio leaves it. In a live session it also gives signals.updated its vocalizations and loudness.
Without the microphone you still get imminence and the subject read, and an audio_absent warning says why the calls are missing.
Body view (optional)
A second camera that sees the person's body, published to ingest.bodyUrl when the session declares inputs.body: true. It corroborates onset, peak and end calls, so more of them arrive with confidence: high. Calls are still made without it. In a live session it also gives signals.updated its contractions and movement.
Stimulation (optional)
If your product controls stimulation, report when it starts, stops or changes with reportStimulation, or a stimulation message on the socket. Imminence from the face alone does not know how long the person has been stimulated; with reports, it does (stimulationAware: true). Report changes, not a heartbeat. A report can also carry the device's level, channels and mode, which every reader receives as stimulation.updated for its display (reporting the device).
When an input goes missing
input.warningevents are raised and cleared as conditions start and stop:face_not_visible,multiple_faces,audio_absent,body_stream_absent,baseline_pending,low_confidence. Each has amessageyou can show the person.- The session's
degradedflag is true while a declared input is missing or the face is out of view, andsession.statusevents report the change. - While the face is out of view, imminence is
null. Treatnullas "unknown", never as "low": hold rather than increase.