Research – Neurolexi Health GmbH

±15 ms

within human inter-rater agreement

9,029

recordings analysed

initial phonemes studied

peer-reviewed publications

Data basis: 9,029 recordings in Swiss and Standard German · 134 healthy and 31 people with aphasia · 16 initial phonemes with ≥ 100 recordings each (Rickert et al. 2026).

How it works

Closer to speech onset — even for weak initial sounds

2×–4× more preciseCloser to speech onset than established tools like Kaldi and Chronset.
±15 ms to referenceWithin the agreement range of human raters.
Phoneme-specificReliably solves the problem of low-energy, voiceless onsets.
No frequency analysisSix optimised features with a reference time and detection window — no heavy spectral analysis needed.

Clinical research

Built on years of peer-reviewed research

2026ARTICLE

Can an initial phoneme-based algorithm improve automatic naming latency detection during picture naming tasks? A feasibility study

Rickert E., Altermatt S. et al. — Biomedical Signal Processing and Control, Vol. 113, Part B, 2026, 108851.

FHNW_NL algorithm, validated on 9,029 recordings (Swiss & Standard German; 134 healthy and 31 people with aphasia): mean deviation from the manual gold-standard measurement within ±15 ms — 2× and 4× smaller than Kaldi and Chronset respectively.

Publication →

2021EMBC

Evaluation of the potential of automatic naming latency detection for different initial phonemes during picture naming task

Park S., Altermatt S. et al. — 43rd Annual Int. Conf. of the IEEE EMBS (EMBC), 2021, pp. 945–950.

Shows a strong positive correlation between automatic and manual latency, and that accuracy depends on the initial phoneme (voiced/voiceless) — the basis for phoneme-specific optimisation.

Publication →

2020EMBEC

Automatic detection of naming latency from aphasia patients – using an extended threshold-based method

Altermatt S., Kuntner K., Rickert E. et al. — Conf. Proc. EMBEC, Nov 2020, p. 71.

First threshold-based method for automatic naming-latency detection; tested on 272 AAT recordings from 8 people with aphasia against the manual gold standard.

Conference paper

2020EMBEC

Patient-friendly speech recognition feedback for aphasia patients

Wyss S., Rickert E., Altermatt S. et al. — Conf. Proc. EMBEC, Nov 2020, p. 283.

Patient-friendly word feedback via speech recognition — cuts the word-error rate from 4.39% to about 2.0% and underpins the live check of the spoken word in the app.

Conference paper

Objective naming latency — scientifically validated.

Closer to speech onset — even for weak initial sounds

Built on years of peer-reviewed research

Research or validate with us?