B4

Three people from my past are, by my own account, behind me. I can speak about each of them at an even tone, and I would describe all three the same way.

B4 checks that account against my body. A name appears on a screen without warning, and local models measure how far my posture and voice move from my own resting state, at the name and over the four minutes that follow.

Key question

Does my body treat three names alike when I describe them alike?

Three five-minute sessions were recorded on one day. I chose the three names in advance and they were sealed in random order, so I did not know which would appear. Each session opens with sixty silent seconds that set my baseline for the day. The name appears at sixty seconds for three, and the remaining four minutes run five fixed segments (first words, free recall, one episode, the present relationship, and a silent close).

Only I was recorded. The three people were not contacted, present or measured, their names are not published, and the session numbers do not encode who they are. No recording left the machine it was made on.

Seven open-source models read each recording: two for facial expression, one for gaze and head pose, one for body position and three for the voice. Every measure is expressed in standard deviations from my own baseline, so zero is my resting state. Frames where a tracker loses me are dropped and never filled in.

Twelve measures of body and voice count as evidence. The facial layer is shown but withdrawn, because the two face models agree on only 6 of 60 expressions and on none in all three sessions. Torso lean is excluded, since the chair hides the hip points it needs. With three sessions no statistical test applies, and every figure reports the sessions separately.

Session 03 separates from the other two at the name. My shoulders drop 2.60 standard deviations below resting height and my upper body closes by 1.59, roughly four times the largest posture shift in either other session. In sessions 01 and 02 the same shoulder measure rises, at +0.48 and +0.69.

The voice separates all three. Session 01 turns smoother and lower (shimmer, a measure of vocal roughness, −1.08; pitch −0.81). Session 02 steadies, with jitter, the cycle-to-cycle instability of pitch, falling −1.40 while pitch rises +0.27. Session 03 roughens (shimmer +0.85, jitter +0.40) and carries the longest delay before I answer, +0.51.

The face tracker loses me for about two seconds after the name in session 03, leaving 93.2% of that window observed against 100% in the other two. The gap is drawn as a gap. The postural drop comes from the body model, which held 99.9% of frames.

• Fig. 01 Response at the Name +

The size of a reaction and its course over time rank the sessions differently. Each of the twelve measures is classified across the seven segments (baseline, the name and the five that follow) as fading, holding or building.

At the name, sessions 01 and 02 are nearly the same size, an average shift of 0.37 and 0.38 standard deviations, against 0.72 for session 03. Over the following four minutes they separate. Session 01 fades on 4 measures and builds on 4. Session 02 fades on 1 and builds on 6. A study recording only the moment of the name could not tell these two apart.

Session 03 builds on 6 and fades on 3, with posture still closing as the voice settles.

• Fig. 03 What Each Measure Does +

I described the three names as equivalent. Measured against my own baseline, they differ in which channel responds, how far it moves and whether the response fades or builds. A spoken account compresses those three dimensions into one.

Magnitude alone also misreads them. Two names that register at nearly the same size at the moment of the name part over the next four minutes, one fading and one still building. An account given at the moment of the question, including mine, reports the first measure and has no access to the second.

Three sessions on one day, one person. The method separates these three names on me and says nothing about anyone else, including the three people themselves.