From the app through the recording to the objective evaluation — initial-phoneme based, precise and without complex frequency analysis.
A short explainer opens the „black box“ of the analysis — and shows why millisecond-precise measurement of speech onset is the real USP.
The patient app is only the surface. The real value is created in one step that most treat as a black box: the millisecond-precise measurement of speech onset.
The app shows an object; the patient names it aloud. Competing apps do exactly this too — a recording is produced. But a recording alone says nothing about language recovery.
At api.neurolexi.ch a speech recogniser confirms the word and provides a rough reference time. The „what“ is solved — the clinically decisive question is the „when“: the naming latency, the objective marker of word retrieval.
But plain speech recognition detects words — not the exact moment the lips open.
Voiceless fricatives like /f/ or /s/ carry almost no energy. Simple tools „miss“ this quiet onset and only trigger at the loud vowel — speech onset is measured too late. In aphasia this distorts every progress measurement.
The weak /f/ frication starts well before the vowel onset.
Instead of costly frequency analysis, the algorithm works purely on the waveform — light enough to one day run live in the app.
In the published study (Rickert, Altermatt et al., 2026), Neurolexi is closer to the manual expert measurement than established tools — and scatters far less.
9,029 recordings · 16 phonemes · schematic depiction of the published results.
Naming latency in milliseconds — a hard, comparable value.
Comparable across sessions — progress becomes verifiable, even at home.
Peer-reviewed with the FHNW, phoneme-specifically trained.
Purely time-based — on the way to real-time measurement in the app.
Four principles shape every measurement — from the patient app to the therapist dashboard.
Swiss hosting and end-to-end encryption protect sensitive patient data.
State-of-the-art models for speech recognition and automatic naming-latency analysis.
Advanced naming-latency detection surfaces deeper insight than manual scoring.
Built on peer-reviewed research and clinical practice with Swiss institutions.
Speech recognition checks the spoken word; the FHNW algorithm measures the response.
From a single patient to scalable therapy success.
To the clinical application →