Facial Action Coding System

46 Action Units.
One universal language.

The Facial Action Coding System (FACS) is the scientific gold standard for facial expression analysis — a taxonomy developed by Paul Ekman and Wallace Friesen that describes every visible movement of the human face through discrete, numbered Action Units (AUs).

EchoDepth tracks all 46 observable Action Units per frame, per person. Because AUs are involuntary — many cannot be consciously controlled — they provide a reliable signal that is independent of self-report.

  • AU1 & AU4: Inner brow raise / brow lowerer — core stress markers
  • AU6 & AU12: Cheek raiser / lip corner pull — genuine vs masked emotion
  • AU17 & AU24: Chin raiser / lip press — suppression and withholding signals
  • Temporal coherence: scored across the full session, not single frames

Example AU activation pattern

AU1 — Inner brow raise0.72
AU4 — Brow lowerer0.64
AU6 — Cheek raiser0.18
AU17 — Chin raiser0.55
AU24 — Lip press0.61

AU pattern consistent with suppressed stress and cognitive load. Session flag: elevated.

Output Model

Three dimensions.
One complete picture.

EchoDepth outputs a continuous VAD score — Valence, Arousal, Dominance — the three-dimensional model of emotional state that underpins modern affective computing.

Valence

V

The positive-to-negative dimension. High valence indicates a positive, comfortable emotional state. Low valence indicates distress, displeasure or anxiety.

In fintech: A valence drop during a specific question in a claims interview may indicate discomfort with that topic.

Arousal

A

The calm-to-excited dimension. Elevated arousal indicates heightened physiological activation — which may reflect stress, urgency, fear or excitement.

In fintech: Sustained high arousal in a mortgage interview correlates with elevated cognitive load — a potential indicator of rehearsed or constructed responses.

Dominance

D

The submissive-to-in-control dimension. High dominance indicates confidence and control. Low dominance indicates a sense of vulnerability or powerlessness.

In fintech: A sudden dominance drop mid-session can indicate the subject has encountered a question they were not prepared for.

Multimodal Analysis

Video is only
the beginning.

EchoDepth combines three signal streams to build the most complete picture of emotional state:

📹
Video / Facial

46-AU FACS analysis at up to 30fps. Temporal coherence scoring across the full session window.

🎙️
Voice / Prosody

Pitch, rate, energy and micro-pause analysis. Detects vocal stress markers independent of content.

💬
Text / Linguistic

Sentiment, hedging, temporal inconsistency and confidence markers in transcribed speech.

Bias Reduction

Trained across
cultures, not just data.

Many emotion AI systems fail outside the demographic of their training data. EchoDepth was deliberately built to avoid this.

  • Training data collected across 6 countries
  • 14 cultural cohorts represented in the model
  • Active bias auditing — cultural expression variance is modelled, not averaged
  • No reliance on posed expression datasets
  • Validated on spontaneous, naturalistic video — not lab conditions

EchoDepth will not be deployed in a context where cultural expression bias would produce discriminatory outcomes. Contact us to discuss responsible deployment.

Privacy by Design

No biometric data stored.
No exceptions.

🔒

No raw video retained

Video is processed in memory. Frames are never stored. Only VAD scores and AU activations are output.

🇬🇧

GDPR compliant

Designed for UK and EU regulatory environments. FCA Regulatory Sandbox participant. Data residency options available.

🏭

On-device processing

Edge deployment option for organisations where video data cannot leave the premises. Full API feature parity.

ISO 9001 infrastructure

Built on Logic Software's ISO 9001 and Cyber Essentials certified infrastructure. NHS and British Airways pedigree.

Ready to go deeper?

Explore the API documentation or talk to the team about a proof of concept.

Developer Docs See Use Cases