Frequency Recognition of Internal Dynamics
Any intelligence that sets out to help a human being understand themselves is born small before the immensity of what it seeks. It will grow exponentially and indefinitely toward becoming the largest and most useful intelligence in aid of those seeking to understand the metaphysics behind the coherent relationship between the soul and its expressions: what the voice — the greatest expression of the human psyche — reveals, the intention behind words, and what the face confesses; and the truth that emerges when the three converge. Because giving humans the means to understand themselves better is, in the end, caring for their very existence.
Overview
Three layers of signal, processed in real time and always compared against an individual 60-second calibration at the start of each session.
Fundamental frequency (F0), cepstral coefficients (MFCC) and their temporal derivatives, zero-crossing rate — extracted continuously and compared to the patient's own baseline.
Tracking of 468 facial landmarks and classification of Action Units (FACS) with onset–apex–offset temporal modeling.
The 12 Perception Zones and two synthesis indices — the Multimodal Power Index (IPM), the "speedometer" of energy and psychomotor vitality, and the Multimodal Deviance Index (IDM), the "compass" of coherence between voice and face — reduce acoustic-facial signals to a visual reading meant to support, not replace, clinical listening.
Audio at 48 kHz, Hamming windowing, FFT run in the browser via the Web Audio API (no DSP dependency on the Python backend — only FastAPI/DuckDB/LLMs for explanation and transcription). Video processed locally at 30+ FPS for the 468 facial landmarks.
All compared against the baseline locked in the first 60 seconds of the session (baseline_mfcc7, baseline_mfcc9, etc.).
Transparency
Unlike most tools in the industry, FROID explicitly separates these two layers — and invites you to check each one.
F0, MFCC, jitter/shimmer, and Action Unit dynamics as signals associated with affective states and psychomotor retardation have decades of published research behind them — see the references.
The 12 Perception Zones, the Facial Dissonance Multiplier, and the IPM/IDM indices are FROID's way of organizing these signals visually. They are not findings from published clinical studies — they are product heuristics, and we treat them as such.
Because mixing the two without warning is the kind of thing that misleads clinicians and harms patients. Any claim of automated diagnosis is false — FROID is a support tool, not a validated diagnostic instrument.
The structure of 12 zones mapped to notes of the chromatic scale is a transposition inspired by commercial voice biofeedback technology, adapted by the FROID team as a visualization interface — it is not published neuroscience.
Who it's for
Psychiatrists, psychologists, and therapists who want an additional layer of objective reading during the appointment — without wasting time on tools that don't integrate with the session.
Frontend in React + Vite + TypeScript (froid-dashboard), FastAPI backend with DuckDB for anonymous analytics and ChromaDB for RAG powering the "FROID Explains" assistant (froid-server). Transcription via GPT-4o-transcribe; explanations via Gemini 1.5 Pro/GPT-4o — always about already-calculated metrics, never generating the acoustic values themselves.
Professional access control per session, with credit consumption per appointment and ownership binding per report (_can_access_report, _professional_access_status).
The differentiator
An LLM-RAG assistant that answers, in clinical language, everything about the session's reading — and cross-references your patient roster with the global anonymous Data-Froid. Comes with 22 ready-made prompts, and you can create your own.
"What does this dissonance mean?", "what does the IPM suggest?" — grounded answers about already-calculated metrics.
"Is this patient above average in risk?", "most similar cases (top 5)?" — the statistical power of thousands of anonymous sessions.
Create prompts in your settings area — and over time, empirically validate the IPM, the IDM, and the other metrics.
Transparency · LGPD
By law, the patient owns their own information. FROID takes this seriously: a dedicated portal, with a login separate from the professional's, where the patient sees and manages everything that is theirs.
Results of completed sessions: key metrics, baseline, segments/summaries, and dissonances — the same reading that supports your professional, now in your hands.
Basic registration data under your control, in line with Brazil's LGPD (Law 13.709/2018) — the right to access and manage your own information in one place.
Download the results of each session individually or all results in a single file — real portability, with no requests and no waiting.
Technical measures applied in the platform to keep your data confidential:
