News

Meta has submitted a patent application for a system that would monitor people’s speech throughout the day, infer their emotional state from how they sound, and record timestamped entries of each assessment. Every emotional read could be anchored to the moment it occurred – including the time, location, surrounding activity and even the way the user was interacting with their phone. The filing describes variants that continuously listen as well as versions that sample audio at scheduled intervals. None of these designs are shipping in a Meta product, and the company has made no consumer announcement; patents often serve to secure an idea long before any product decision is made.

The application, listed as US 2026/0182881, was submitted by Meta Platforms in December 2025 and published on July 2, 2026. It names Lachlan Dunn as the sole inventor and claims priority to a provisional submission from December 2024. The patent-analysis site Patentlyze was the first to highlight the document.

Although the filing’s title links emotional-state inference with a real-time fitness coach, the claims show the emotional-analysis component is central. Of the 20 claims, three independent claims focus solely on emotion detection, while workout-related features appear only as dependent claims that build on those emotion-detection assertions.

How the system would work: a device – whether smart glasses, a smartphone, a smartwatch, headphones or a smart speaker – records a user’s speech during the day. The audio is transcribed and an AI model trained to infer mood analyzes both what is said and how it is said: vocal tone, speaking rate, a sigh, laughter and similar prosodic cues. Each segment of speech receives an emotional label, the surrounding context is associated with that label, and over a defined window – such as a day or month – the system compiles a summary of the user’s emotional patterns.

The filing emphasizes traceability: the system can link every emotional assessment back to the actual words that produced it, what the patent calls a citation. In one illustrative example, an “anger” reading is accompanied by the precise sharp language that led to that conclusion.

One diagram follows an individual through a typical day: quiet speech during a morning video meeting at home, laughter with a friend at dinner, and a 9:15 p.m. sigh captured by a smart speaker. In that illustration, the speech events are “time stamped and logged on servers,” and the system would present the user with an aggregated summary like a short report noting that they tend to sigh most often before bed, feel happiest in social company, and have expressed more gratitude than usual this month.

The proposed system extends beyond audio. The patent describes incorporating biometric and eye-tracking inputs – pupil diameter, blink frequency and even eye moisture – to detect stress or tears. It would also monitor device behaviors such as the posts a user views or likes, total screen time and the speed of switching between apps. All signals would feed into a unified emotional profile.

The filing’s other major strand is a mood-aware workout coach. In that scenario, smart glasses observe a user’s form in a mirror and provide real-time guidance – for example, advising a deeper squat and then encouraging additional reps. This coaching component also factors in mood: if the system detects fatigue or discouragement, it may scale back intensity; if it judges the user to have surplus energy and to be slacking, it may chastise or push them. The patent asserts that no human trainer could match the system’s continual precision or operate at that scale throughout the day.

This concept is not entirely new. Amazon introduced mood-reading via voice with its Halo wearable in 2020: the Halo “Tone” feature analyzed pitch and speaking pace to characterize how users sounded during the day – calm, frustrated and similar states – and processed samples locally on the phone before deleting them rather than uploading them to the cloud. That feature nevertheless attracted scrutiny: in December 2020 Senator Amy Klobuchar pressed federal health regulators about Halo’s collection of voice-tone and body-scan data, calling the approach unusually invasive. Amazon discontinued the Halo line in 2023, though it did not explicitly link the shutdown to privacy concerns.

The crucial difference in Meta’s filing is breadth rather than just storage location. Some embodiments keep processing on-device, while others send logs to servers; more importantly, where Halo relied solely on voice, Meta’s design would combine audio with eye-tracking and phone-use telemetry. Regulators have begun questioning whether such emotion-inference methods are scientifically robust, and they are drawing legal boundaries.

Since February 2025 the EU’s AI Act has prohibited AI systems that infer people’s emotions in workplaces and educational settings except for narrowly defined medical or safety reasons, exposing violators to fines up to €35 million or 7% of global annual turnover, whichever is higher. The Act’s drafters explicitly warned that emotional expression differs between individuals, cultures and moments, undercutting the reliability of blanket inferences. That ban does not apply across-the-board to consumer products, however. A separate EU rule scheduled for August 2026 will require systems that infer emotions from biometric signals to disclose that capability. Whether a voice-first coaching tool falls under those disclosure requirements is debatable, but a system that also analyzes pupils and blink rates would more clearly be captured by the biometric rule.

The workout coach is only one envisioned application. The underlying idea is a continuous log of emotional readings tied to time, place and activity. Amazon’s mood-reading tool relied solely on voice and was removed in 2023; Meta’s filing imagines extending that reach into eye metrics and device behavior. For now, the proposal remains a patent on paper – the only thing keeping it out of daily life is that nobody has actually built it.