Automatic evaluation of spontaneous speech samples of the Aachen Aphasia Test

Automatic evaluation of spontaneous speech samples of the Aachen Aphasia Test - autoAAT

Status

Active, not recruiting

Phases

Unknown

Study type

Observational

Source

DRKS

Registry ID

DRKS00031959

Enrollment

800

Registered

2024-08-15

Start date

2024-06-04

Completion date

Unknown

Last updated

2026-01-12

For informational purposes only — not medical advice. Sourced from public registries and may not reflect the latest updates. Terms

Conditions

R47.0

Interventions

Group 1: Development and testing of automated procedures for the transcription and evaluation of spontaneous speech interviews of patients with aphasia collected with the Aachen Aphasia Test (AAT). T

Eligibility

Sex/Gender

All

Inclusion criteria

Inclusion criteria: Presence of aphasia/aphasic residual disorder regardless of etiology, e.g. vascular (bleeding, infarct), traumatic brain injury, degenerative (primary progressive aphasia), inflammatory (multiple sclerosis), etc. German at native language level (pre-onset) Treatment at the aachen aphasia ward

Exclusion criteria

Exclusion criteria: none

Design outcomes

Primary

Measure	Time frame
The aims of the study are quality and feasibility of standardized transcription of aphasic speech interviews by automated procedures. Additionally, the evaluation of the accuracy of the diagnostic machine learning model in predicting different types of aphasia from the spontaneous speech scores of the Aachen Aphasia Test (AAT). The following metrics are used for this purpose: •Word-Error-Rate (WER) •F1-Macro-Score •Area-Under-the-Curve (AUC) •Unweighted Average Recall (UAR) As this is not an experimental study but an optimization process, the transcription is checked iteratively with the goal of continuous improvement. The testing is performed automatically by comparing the model's output with the manual transcription by the therapist. During this comparison, the above-mentioned metrics are calculated. These metrics enable a comprehensive evaluation, especially for unbalanced data, as they are less susceptible to data imbalance.	—

Measure

Time frame

The aims of the study are quality and feasibility of standardized transcription of aphasic speech interviews by automated procedures. Additionally, the evaluation of the accuracy of the diagnostic machine learning model in predicting different types of aphasia from the spontaneous speech scores of the Aachen Aphasia Test (AAT). The following metrics are used for this purpose: •Word-Error-Rate (WER) •F1-Macro-Score •Area-Under-the-Curve (AUC) •Unweighted Average Recall (UAR) As this is not an experimental study but an optimization process, the transcription is checked iteratively with the goal of continuous improvement. The testing is performed automatically by comparing the model's output with the manual transcription by the therapist. During this comparison, the above-mentioned metrics are calculated. These metrics enable a comprehensive evaluation, especially for unbalanced data, as they are less susceptible to data imbalance.

—

Secondary

Measure	Time frame
Additional metrics (e.g., accuracy, precision) may be used if necessary.	—

Countries

Germany

Contacts

Public ContactDorothea Peitz

Klinik für Neurologie der Uniklinik RWTH Aachen

dpeitz@ukaachen.de+49 241 8089968

Outcome results

None listed

Source: DRKS (via WHO ICTRP) · Data processed: Feb 4, 2026