CoChat

Case Report: Tailored automatic speech recognition in global aphasia with dysarthria - a single case proof of concept

Open access

The DOI matches its Crossref record; 21 of the 21 references it cites were found in a registry or an index; 5 of the 9 sources its claims cite were checked for support.

Abstract

Background: Severe dysarthria and global aphasia drastically reduce speech intelligibility, confining communication to familiar partners. Automatic speech recognition (ASR) systems may show limited performance when processing such atypical speech. Objective: To determine whether a speaker-dependent Voice-Input Voice-Output Communication Aid (VIVOCA) embedded in the CapisciAMe app can decode the speech of a person with severe dysarthria and aphasia more accurately than rehabilitation professionals human listeners (RPHL). Methods: We conducted a single-case proof-of-concept study. A 34-year-old woman, 15 years post-stroke, recorded 1,120 utterances of 13 target-words across five prompting modalities. A compact convolutional neural network (cnn-trad-fpool3) was trained on these samples and evaluated on an independent set of 936 utterances. Intelligibility was benchmarked against 12 RPHL familiar with the patient. The primary outcome was word-level accuracy. Results: The tailored ASR achieved 72.65 % accuracy, outperforming familiar RPHL (mean = 56.75 %, SD = 12.91). Conclusions: A personalized ASR system can exceed the intelligibility of human listeners for profoundly disordered speech, supporting its use as an assistive communication technology.

Overview

CoChat Summary

## TL;DR

This single-case proof-of-concept found that a speaker-dependent ASR system embedded in the CapisciAMe app decoded one woman’s severely dysarthric/global-aphasic speech better than familiar rehabilitation professionals. On the held-out test set, ASR accuracy was 72.65% versus a human mean of 56.75%.

## Key findings

  • The tailored ASR correctly identified 680 of 936 test utterances, for 72.65% word-level accuracy with 95% CI 69.71–75.41.
  • The 12 familiar rehabilitation professionals achieved mean accuracy = 56.75% (SD = 12.91; bootstrap 95% CI 49.38–63.48), with individual accuracy ranging from 38.5% to 73.1%.
  • The ASR outperformed the listener group by 15.9 percentage points; this difference was statistically significant by both one-sample t-test: t(11) = −4.27, p = 0.001 and Wilcoxon signed-rank test: W = 1.0, p = 0.001.
  • ASR performance varied by prompting modality, from 54.6% for picture-only prompting (M3) to 84.1% for audio-only prompting (M2).

## Methods

  • Design: Single-case proof-of-concept study comparing a personalized speaker-dependent ASR system to human intelligibility judgments.
  • Sample/data: One 34-year-old woman, 15 years post-stroke, with global aphasia and severe dysarthria; 1,120 utterances were used for training/model development and 936 held-out utterances for testing. Human comparison used 12 Italian rehabilitation professionals.
  • Analysis: A compact CNN (cnn-trad-fpool3) was trained on MFCC-based features; ASR accuracy was evaluated on a session-based held-out test set. Human listener accuracy was summarized descriptively, with bootstrap CIs, and ASR vs human performance was compared using a one-sample t-test and Wilcoxon signed-rank test.

## Limitations

  • The paper explicitly notes that this is a single-case design, so the findings are not immediately generalizable to broader populations.
  • The limited training vocabulary constrained communicative breadth, according to the paper.
  • The authors state that the human benchmark was a familiar-listener baseline (rehabilitation professionals with ongoing exposure to the patient), so the ASR advantage should not be generalized to the broader population of human listeners.
  • Evident from the text: there was no prospectively defined shared evaluation subset for a fully paired item-level comparison between ASR and human responses.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3389/fresc.2026.1813312

Cites

21 of 21 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 Axial brain MRI scans highlight an extensive neuroanatomical disruption affecting the left opercular region of the patient. The figure shows T2-weighted MRI with reversed grayscale contrast performed 18 months from the cerebrovascular accident. The first image series from left to right shows progressive slices of the brain at different levels from top to bottom. The localization and extent of the lesion was depicted on the left hemisphere with reference to Broadman's map areas. The most posterior part of the superior and middle temporal gyri was partially spared as well as the angular gyrus and part of the supramarginal gyrus.
Figure 1 Axial brain MRI scans highlight an extensive neuroanatomical disruption affecting the left opercular region of the patient. The figure shows T2-weighted MRI with reversed grayscale contrast performed 18 months from the cerebrovascular accident. The first image series from left to right shows progressive slices of the brain at different levels from top to bottom. The localization and extent of the lesion was depicted on the left hemisphere with reference to Broadman's map areas. The most posterior part of the superior and middle temporal gyri was partially spared as well as the angular gyrus and part of the supramarginal gyrus.CC BY · PubMed Central
Figure 2 Mean word-level accuracy for the tailored ASR system and the human listener control group. For the ASR, error bars represent 95% confidence intervals (Wilson score interval). For the human listener group, error bars represent the standard deviation (SD); black dots indicate individual listener scores (n = 12). The ASR correctly recognized 680 of 936 utterances (72.65%, 95% CI 69.71–75.41); the listener group mean was 56.75% (SD = 12.91%).
Figure 2 Mean word-level accuracy for the tailored ASR system and the human listener control group. For the ASR, error bars represent 95% confidence intervals (Wilson score interval). For the human listener group, error bars represent the standard deviation (SD); black dots indicate individual listener scores (n = 12). The ASR correctly recognized 680 of 936 utterances (72.65%, 95% CI 69.71–75.41); the listener group mean was 56.75% (SD = 12.91%).CC BY · PubMed Central

Verification

Claims the paper makes

5 Valid · 0 Not valid · 3 Not checked · 39 Own results and interpretations

12 more claims

Create a free account to see all 47 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 18
  • Probably not a substantial claim. 17

Create a free account to see these claims.

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Consistentt(11) = -4.27reported p = .001 · recomputed p = .001

    “t(11) = −4.27, p = .001”

1 of 1

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Self-check results appear here once an independent check has confirmed them.

Citations

Cites

Show all 21 references

Research works printed but not linked: 3

  1. Jayaraman, 2025, Dysarthria
  2. Whitehill, 2010, Environmental barriers to communication for individuals with dysarthria, J Med Speech Lang Pathol, 18, 141
  3. Capasso, 2001, Esame Neuropsicologico per L’afasia (E.N.P.A.)

Metadata

Authors
Mulfari, D., Cardile, D., Campana, S., Voci, L., Vicario, C. M., Calabrò, R. S., Corallo, F., Mulfari, S., & Tomaiuolo, F.
Affiliations
  • University of Messina (IT)
  • Magna Graecia University (IT)
  • Neurorehabilitation Unit, Auxilium Vitae Volterra
  • Centro Neurolesi Bonino Pulejo (IT)
Corresponding author
Cardile, Davide
Venue
Frontiers in Rehabilitation Sciences
Published
2026-06-03
Volume
7
Pages
1813312-1813312
ISSN
2673-6861 (ISSN-L)
OpenAlex
W7163425723
Cited by
0
References
21
FWCI
0.00
Type
journal article
Language
en
Open access
Locations
  • Frontiers in Rehabilitation Sciences (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
Università degli Studi di Messina
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Speech Recognition and Synthesis
  • Health Sciences › Health Professions › Occupational Therapy › Assistive Technology in Communication and Mobility
  • Health Sciences › Health Professions › Speech and Hearing › Voice and Speech Disorders
Keywords
global aphasia, dysarthria, speech recognition, assistive communication technology, convolutional neural networks, stroke, speech intelligibility, personalized speech recognition

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-01, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-01 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Case Report: Tailored automatic speech recognition in global aphasia with dysarthria - a single case proof of concept. https://app.cochat.ai/papers/10.3389/fresc.2026.1813312
The paper
Mulfari, D., Cardile, D., Campana, S., Voci, L., Vicario, C. M., Calabrò, R. S., Corallo, F., Mulfari, S., & Tomaiuolo, F. (2026). Case Report: Tailored automatic speech recognition in global aphasia with dysarthria - a single case proof of concept. Frontiers in Rehabilitation Sciences. https://doi.org/10.3389/fresc.2026.1813312

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.