TELL — Micro-Expression Lie & Interest Detector
Summary
TELL is a real-time overlay for video calls and AR glasses that tracks 43 facial action units to detect micro-expressions, surfacing indicators of deception, boredom, and attraction. It builds a personal baseline and analyzes facial landmarks and muscle movements to provide calm, contextual pop-up signals on screen, empowering users to interpret non-verbal cues in real-time for various applications like negotiations, dating, and interviews.
Key Insights
It decodes 43 muscle movements inspired by FACS for specific emotions.
TELL employs a model inspired by the Facial Action Coding System (FACS) to interpret the movements of 43 facial muscles. For instance, the simultaneous flashing of AU4 (lip corner tightening) and AU7 (lid tightening) can indicate concealed anger, while an asymmetry in AU12 (lip corner pull) suggests politeness rather than genuine amusement.
TELL surfaces signals, it does not read minds or provide definitive verdicts.
TELL is designed to highlight statistically significant facial signals against a personal baseline, acting as a spotlight on potential cues rather than a mind-reading tool or a definitive judgment system. Users are cautioned against trusting anyone claiming otherwise.
Sections
Introduction to TELL
TELL is a real-time overlay for video calls and AR glasses.
TELL functions as a real-time overlay designed for use with video calls and augmented reality glasses.
It tracks 43 facial action units to surface emotions and states.
The system tracks 43 distinct facial action units, which are the fundamental movements of facial muscles, to identify and surface indicators of deception, boredom, and attraction.
TELL displays calm pop-ups on screen with analysed data.
Instead of overwhelming the user, TELL presents the analysed information as calm, discreet pop-up notifications on the edge of the screen.
Key metrics include 43 facial action units, 118ms latency, and 91.4% baseline accuracy.
The system boasts a technical performance with 43 facial action units tracked, an end-to-end latency of 118 milliseconds, and a baseline accuracy rate of 91.4%.
How TELL Reads a Face
TELL tracks 68 facial landmarks at 30fps to build a neutral baseline.
The system accurately tracks 68 distinct facial landmarks in real-time at a rate of 30 frames per second. It establishes a personalized neutral baseline during the first six seconds of interaction, against which all subsequent micro-shifts in facial expression are measured.
It decodes 43 muscle movements inspired by FACS for specific emotions.
TELL employs a model inspired by the Facial Action Coding System (FACS) to interpret the movements of 43 facial muscles. For instance, the simultaneous flashing of AU4 (lip corner tightening) and AU7 (lid tightening) can indicate concealed anger, while an asymmetry in AU12 (lip corner pull) suggests politeness rather than genuine amusement.
Cue synthesis provides calm, one-line pop-ups with signal confidence.
The system synthesizes the analyzed cues into a concise, single-line pop-up message that appears at the edge of the user's view. These messages indicate what specific facial shift occurred, when it happened, and the system's confidence level in the interpretation, emphasizing presentation of signal rather than definitive verdicts.
Applications and Benefits
In negotiations, it detects deception by comparing verbal claims with facial cues.
TELL monitors the live confidence delta between a person's spoken words and their facial expressions. When a discrepancy arises, such as a tightening of the lip corner after claiming certainty, the indicator (e.g., a confidence bar) changes color, signaling a potential bluff. Pilot studies showed a +37% increase in closed deals.
In dating, it identifies attraction signals like pupil dilation and lip pressing.
For dating scenarios, TELL detects genuine signals of attraction, including pupil dilation, lip pressing, and cheek raises. It can notify the user when interest levels spike, helping to reduce uncertainty. Self-reported data indicated a 3.4x increase in second dates.
In interviews, it spots disengagement through micro-yawns and gaze drift.
TELL helps identify disengagement in interviews by detecting subtle cues like micro-yawns, gaze aversion, and small head tilts away from the camera. An alert system can signal these shifts, prompting the interviewer to re-engage the panelist. This led to a 52% reduction in dropped calls compared to a baseline without the overlay.
It integrates seamlessly with AR glasses for an invisible, in-person experience.
TELL is compatible with AR glasses like Ray-Ban Meta and XREAL Air 2 Ultra, providing an in-person feel where cues appear only when the signal is strong, maintaining a clean field of view. The system operates with 118ms latency and is designed to be eye-safe and opt-in only.
Real-Time Example: A Moment Decoded
Example shows detecting micro-tension when claiming alignment.
During a video call, when a speaker claims 'we're totally aligned on the timeline,' TELL detects a micro-tension in the lip corner 0.4 seconds after the word 'aligned,' contradicting the stated confidence.
Detects gaze aversion during specificity pressure in conversation.
Following a question about a locked end-of-quarter deadline, the other speaker exhibits 2.9 seconds of downward gaze, a common indicator when faced with pressure for specificity.
Identifies recovery and honesty score increase after partial admission.
When the speaker admits '...mostly. There's a couple of moving pieces,' TELL detects a recovery in facial cues, with the honesty score beginning to climb again as baseline is restored.
Provides real-time scores for Interest, Deception Risk, Attraction, and Boredom.
Throughout the interaction, TELL continuously updates scores for various metrics including Interest (shown at 82), Deception Risk (at 41), Attraction (at 64), and Boredom (at 12), providing a dynamic view of the other person's state relative to a baseline.
Hardware Recommendations
Webcam quality significantly impacts TELL's performance.
The effectiveness of TELL's analysis is directly influenced by the resolution and low-light performance of the webcam used. Higher quality webcams provide a cleaner signal.
Recommended webcams offer specific advantages for facial tracking.
The Logitech MX Brio 4K is recommended for its clean low-light frames and fast face mesh locking. The Insta360 Link 2C is advised for stand-ups due to its AI gimbal maintaining face centering. The Opal Tadpole is suggested for travel as a pocketable, road-ready option impacting negotiations.
Foundational Knowledge & Resources
TELL's cue library is informed by key non-verbal communication principles.
The design of TELL's cue library is heavily influenced by established research and principles in non-verbal communication, including tactical empathy, body language interpretation, emotional intelligence, and negotiation tactics.
Recommended reading augments TELL's capabilities with deeper understanding.
Complementary reading suggested includes 'Never Split the Difference' for tactical empathy, 'What Every BODY is Saying' for detailed non-verbal cues, 'Emotional Intelligence 2.0' for improving EQ, and 'Getting to Yes' for principled negotiation frameworks, which together enhance the user's ability to act on TELL's signals.
Clarifications and Disclaimers
TELL surfaces signals, it does not read minds or provide definitive verdicts.
TELL is designed to highlight statistically significant facial signals against a personal baseline, acting as a spotlight on potential cues rather than a mind-reading tool or a definitive judgment system. Users are cautioned against trusting anyone claiming otherwise.
Analysis is performed locally; no data is recorded or transmitted.
All facial analysis is conducted locally on the user's device. No video frames or facial data leave the machine, ensuring privacy as no recording occurs and the other party is unaware of the analysis.
Consent is prioritized, with local analysis and an easy opt-out feature.
The system emphasizes consent by running analysis locally and including a one-tap 'off' feature. TELL avoids operation in jurisdictions where covert emotion analysis is regulated.
TELL supports specific operating systems and popular communication platforms.
Currently, TELL is available for macOS 14+ and Windows 11, and integrates with platforms such as Zoom, Google Meet, Microsoft Teams, FaceTime, and Discord, as well as AR glasses like Ray-Ban Meta and XREAL Air 2 Ultra via a companion app.
Ask a Question
*Uses 1 Wisdom coin from your coin balance
Watch / Source Article
TELL — Micro-Expression Lie & Interest Detector
Read the original web article directly on read-the-room-why-dont-ya.lovable.app
Open Web Note Link
