Follow the reasoning one question at a time, from how the brain creates a feeling to how a device could measure one.
It treats the brain as a system you can model, like any other information-processing machine. Its central idea: the brain isn't a passive receiver of the world, it's a prediction engine, constantly forecasting what comes next and correcting itself. That one principle explains a surprising amount of how we think, move, and feel.
The brain predicts everything, including what's happening inside your own body: heartbeat, breath, energy. It forecasts your internal state and checks the forecast against reality. Emotion is the brain's reading of those bodily signals. Feeling isn't lifted straight off the body, it's the brain's best explanation of what the body is doing. This is the mainstream view now, not a fringe one.
The brain doesn't just sense the body; it acts to keep it on track, anticipating what you'll need and adjusting before you need it. That's active inference: prediction that drives action, not just observation. Applied to the body, this regulation is called allostasis, or body-budgeting, and it runs against a personal baseline: each of us has a range our body expects to stay within. Your state, in this view, is simply where you sit relative to your own baseline, exactly the kind of thing you can measure.
Yes, and the field is decades old. It's called affective computing, founded at MIT, alongside a mature body of work reading emotion from the voice. Your physiology and vocal tone carry real, measurable signals: heart-rate patterns, vocal strain, and speech rhythm all shift with your state. The question was never whether the signal exists; the hard part is reading enough of it, in context, against the right baseline.
Then you've built the thing the science has been pointing toward. Continuous, real-world signals (body, voice, and context together, all day) are far richer than the brief lab snapshots most research leans on. One model, trained across everyone, learns how signals map to states in general; it then reads each person against their own baseline, because the same signal means different things for different people: one person's resting heart rate is another's stressed one. The model never changes from person to person; what changes is the baseline it reads you against, learned from your own data over time. Do that across thousands of people and you have what the field has never had: the data to learn how feeling actually looks in real life.
You can't read it from any single signal, so you triangulate three. Your body shows how activated you are: heart-rhythm patterns that separate stress from calm. Your voice adds the colour: expressive and bright versus flat or strained. And your context sets the odds: the app you're in, how calm your surroundings are. Any one alone is ambiguous; fused together, against your own baseline, they resolve into a read of positive or negative that no single signal could give.
You teach it once. In an early training phase, people mark how they feel in the moment, so the model learns to map real signals to real states from first-hand labels, not guesses. A controlled lab study, using proven methods for reliably evoking specific states, anchors and validates that mapping. After that, the model reads your state on its own: the in-the-moment labeling is the scaffolding used to build the model, not something you do forever.
Every piece here is established science: the brain as a predictive, body-regulating system; emotion as its reading of the body; decades of decoding affect from physiology and voice; the power of reading each person against their own baseline. What's new isn't any single part. It's bringing them together, at scale, in the real world: the next step the science has been working toward. That's what Anoria is building.