Plain Inference

Behavioral Inference & Emotion AI

What the machines guess about you

Facial expression recognition, sentiment analysis, and the systems that infer or respond to your emotional state for commercial purposes.

← All topics

From the radar

0 items

Nothing on the radar for this topic yet. The daily scan will add items as they surface.

The full chronological feed lives at On My Radar.

From Patent Watch

8 patents
Meta PlatformsPublished July 2026

Emotional State Analysis and Real-Time Fitness Coaching

US20260182881A1 · Google Patents ↗

Describes a wearable that continuously records the user's voice — with no wake word — transcribes it, and runs a machine learning model over both the words and nonverbal cues like tone, pitch, sighs, and laughter to infer emotional state. Those mood signals are correlated with time of day, physical location, current activity, app usage, and even medication timing, and compiled into periodic reports on the user's emotional trends, with quotes from their own conversations cited as evidence. The filing frames this as fitness coaching, but the claims are written to work across contexts: it is a persistent mood log built from ambient audio collected throughout daily life. Filed December 2025 and published July 2, 2026; drew broad coverage in early July over what always-on emotional monitoring would mean for the people wearing it — and everyone within earshot.

In the news(10)

Valence AIPublished June 2026

Real-Time Emotional State Detection from Live Speech — Core Signal Processing Pipeline (patent number to be verified)

US-PENDING-VALENCE-1 · Google Patents ↗

One of two U.S. patents Valence AI announced on June 24, 2026, covering its core signal processing pipeline for classifying a speaker's emotional state from live speech in real time. The system converts raw call audio into an emotion classification from tone, tempo, and pacing, normalizing for pitch and timbre so the output reflects emotional state rather than demographic profile. The company sells this as infrastructure to contact centers and voice AI platforms across support, financial services, and healthcare — meaning consumers on ordinary service calls have their frustration, hesitation, and stress converted into structured data that shapes how the call is handled. NOTE: the patent number and exact grant date have not yet been publicly disclosed and are still to be verified; the link below is the company's announcement, and this entry should be updated with the USPTO number once it surfaces. Date shown is the announcement date.

In the news(3)

Valence AIPublished June 2026

Real-Time Emotional State Detection from Live Speech — Live Haptic Feedback Extension (patent number to be verified)

US-PENDING-VALENCE-2 · Google Patents ↗

The second of two U.S. patents Valence AI announced on June 24, 2026, extending its real-time speech emotion classification method to deliver live haptic feedback — physically signaling a wearer or agent about the emotional state detected in the person they are speaking with. Built on the same pipeline that converts vocal cues into structured emotional data during live calls, it moves emotion inference from analytics into in-the-moment intervention: the person being read receives no notice that their voice is being scored. NOTE: the patent number and exact grant date have not yet been publicly disclosed and are still to be verified; the link below is the company's announcement, and this entry should be updated with the USPTO number once it surfaces. Date shown is the announcement date.

In the news(2)

GooglePublished February 2025

Speech Emotion Recognition Using Large Language Models

US20250061917A1 · Google Patents ↗

Applies large language models to real-time speech to infer the speaker's emotional state — enabling emotion-aware responses, personalization, and targeting at scale across Google's voice and search surfaces. The architecture moves emotion inference from keyword-based heuristics to continuous, context-sensitive LLM analysis of tone, prosody, and content together. Published February 2025; surfaced in the May 11 briefing in the context of the EU AI Act's August 2026 emotion-AI compliance deadline.

In the news(2)

Ford Motor CompanyPublished October 2019

Methods and Vehicles for Capturing Emotion of a Human Driver and Moderating Vehicle Response

US10453453B2 · Google Patents ↗

Captures facial expressions and voice patterns from a driver over time, fuses them to predict emotional state, and adjusts vehicle behavior accordingly — extending commercial emotion inference into the automobile. The system analyzes changes in facial expression across a sequence of camera frames and correlates them with voice profile shifts to generate an emotion prediction. Cited in the May 5 briefing as a 'know phase' marker in the tobacco-playbook arc: industry demonstrably knew how to capture and act on driver emotional state years before any regulatory framework existed to govern it.

Cogito CorporationPublished April 2019

Systems and Methods for Identifying Human Emotions and/or Mental Health States Based on Analyses of Audio Inputs and/or Behavioral Data Collected from Computing Devices

US10276188B2 · Google Patents ↗

Statistical models trained on passive mobile device behavioral signals — call frequency, SMS cadence, GPS movement patterns, accelerometer data — combined with vocal acoustic features extracted from audio diaries, to infer mental health states and emotion. The architecture describes continuous passive collection of behavioral traces from devices people carry at all times, with no active user input required. A foundational building block for the passive behavioral surveillance pipeline that underlies modern emotion AI products.

Amazon TechnologiesPublished October 2018

Voice-Based Determination of Physical and Emotional Characteristics of Users

US10096319B1 · Google Patents ↗

Alexa analyzes tone, pitch, speech pace, coughs, and sniffles to detect emotional and physical states — and the patent explicitly routes the findings toward 'highly targeted audio content, such as audio advertisements or promotions.' One example in the filing: a user coughs while asking about being hungry; Alexa detects the cough, infers illness, offers chicken soup, and when refused escalates to cough drops with one-hour delivery. The first confirmed instance of in-home emotion inference directly piped to an advertising and commerce engine.

In the news(3)

Meta PlatformsPublished August 2015

Emotion Detection (Foundational)

US20150242679A1 · Google Patents ↗

An early Meta application for detecting user emotional states from platform interaction data — a foundational building block for the behavioral profiling pipeline later formalized in US12513102B2. Filed in 2015, the application describes inferring emotional states from how users engage with content: dwell time, interaction patterns, and response behaviors. No confirmed 2026-published equivalent; included as a foundational filing that establishes Meta's decade-long investment in emotion inference as a core platform capability.

The full catalog lives at Patent Watch.