Every changehas a moment.
Speech, voice, face and movement are measured against the speaker’s own range. Each reading links to the recording.
A recorded conversation, mapped from start to finish.
Run locally. Export the data.
Start withthe speaker’s range.
The soft ring is built from stretches of this recording.
Then watchwhat moves.
Read clockwise from the top, start to finish. Each fine line is one sampled moment. The longer the line, the bigger the change.
Give the changeits context.
The short arcs inside the ring are answers. A longer arc is a longer answer, so every change stays inside the exchange it came from.
Both instruments,at once.
Gold marks the answers where the camera and the microphone both rose above the speaker’s range at the same time.
Follow the change.Inspect the measurement.
Each exchange links its measurements to the speaker’s own range and to the moment in the recording they came from.
What else can we observe?
Speech pace, pauses, voice, face and body are aligned to the same exchange. Their timing shows which changes occur together.
What we measure
-
01
Speech timing
- speech rate
- pause burden
- response latency
-
02
The face
- 52 blendshapes
- 15 action families
- 6 left–right asymmetry
- microexpression
- eye aperture
- gaze
- head
-
03
The voice
- pitch
- energy
- jitter
- shimmer
- harmonicity
- spectrum
-
04
The body
- shoulder tilt
- width
- height
- asymmetry
- head-over-torso
How we read it
- Shared time
- Face, voice and body land on one 100-millisecond grid — anchored to words, turns and exchanges.
- Whose change
- One person is locked in as the subject. Mouth movement and the microphone together decide whose turn is whose.
A change from what?
People speak and move differently. We pool stretches of a person’s own recording to build a separate range for face, voice and words. Each range covers how much that person varies from one moment to the next.
Speaker Atheir range · 120–163
190a departure
an ordinary moment
Speaker Btheir range · 155–200
Each hairline counts samples at that rate, in words per minute; the light band marks the middle 80% of a speaker’s samples. In this example, 190 words per minute falls outside Speaker A’s range and inside Speaker B’s: the same rate is a change for one speaker and an ordinary moment for the other.
How the range is built
Face, voice and words each get their own set of windows from the same recording.
A window qualifies when it holds enough clear observations of the person and a steady pattern inside it. Each range is built from at least two qualifying windows, so it carries the variation between them.
The instrument.
Open to inspection.
The Viewer places the recording, transcript and measured changes on the same timeline.
The examples on this page come from previously analysed recordings. PHOSPHOROS can run locally on your own footage.
The Board opens one exchange in detail, with its readings and source moment. The measurements can be exported.
The evidence canleave the interface.
Export includes a readable report, measurements, coverage and provenance.
The instrument ranks changes; the evidence shows what was measured.
Open the report and file details
## Analysis contract
This package records exchange-level measurements of voice
and face against the subject’s own recorded range,
with the exchanges the instrument ranked most
departed first. Read it for what the channels did and
where they agreed.-
measurements.csv 12 columns
One row per bin, per channel, per exchange.
-
timeseries.jsonl 4 fields
One record per exchange, including whether measurements were missing, unavailable or excluded.
-
manifest.json 17 keys
Schema, provenance, coverage and checksums for the other three.
Let’s talk.
If this interests you as a tool, an idea, or something we could build on, I’d like to hear from you.
- hello@phosphoros.xyz