Early Detection of ALS in Absence of Speech Impairments With Computer Audition

MCML Authors

Adria Mallol-Ragolta

→ Group Björn Schuller
 Health Informatics

Björn Schuller

Prof. Dr.

Principal Investigator

Health Informatics

Abstract

We investigate whether Amyotrophic Lateral Sclerosis (ALS) can be detected in patients without speech impairments utilising computer audition techniques. We exploit the information embedded in the patients’ speech while performing five different speech tasks. Specifically, producing the sustained vowel /a:/, repeating the syllables /da/-/da/ and /da/-/ba/ (separately), reading a text passage, and describing a picture. The implemented models are task-dedicated, as they are solely trained and assessed with the speech samples of the corresponding task. We conduct our experiments on the novel, German-speaking AIMnd dataset. We define the Unweighted Average Recall (UAR) as the evaluation metric. When differentiating ALS patients with normal speech from controls – binary classification –, the best models, which obtain a UAR score of 88% on the Test set, mostly exploit the speech samples corresponding to the /da/-/ba/ task. When including the ALS patients with, at least, detectable speech disturbances in the detection – three-class classification –, the best model on the Test set scores a UAR of 70%, also exploiting the speech samples corresponding to the /da/-/ba/ task.

inproceedings

---

AIME 2025

23rd International Conference on Artificial Intelligence in Medicine. Pavia, Italy, Jun 23-26, 2025.

---

Authors

A. Mallol-Ragolta • M. Gonzalez-Machorro • R. von Heynitz • K. Scherzer • I. Cordts • B. W. Schuller

---

Links

DOI

---

Research Area

 B3 | Multimodal Perception