*Presented at the NEALS Conference 2026

Introduction: Traditional scales like the ALSFRS-R lack the sensitivity to detect early bulbar involvement in spinal-onset ALS. Automated speech analysis provides an objective, non-invasive tool to capture subclinical micro-changes before clinical manifestation.

Objectives

• Determine if automated speech metrics can identify early bulbar involvement in spinal-onset ALS

• Evaluate the relationship between speech metrics and respiratory function (FVC%)

Methods:

Data Collection

• In-clinic smartphone recordings (Vox4Health [1], VOC-ALS data set [2]) and forced vital capacity (FVC%)

• Speech tasks 

  • DDK (pa-pa-pa, ta-ta-ta, ka-ka-ka)

  • Sustained vowel phonation (i, a, u)

Data Processing

• Acoustic speech feature extraction via SIGMA, ki:elements’ speech processing pipeline

• Composite scores using DDK features [3]

• Maximum phonation time of vowels

Results- DDK Composites:

Speed vs. Precision: DDK rate remained normal, but articulatory stability was significantly impaired across all syllables (all p < 0.02).

Velar Sensitivity: Articulation precision was specifically reduced for syllable /ka/ (p = 0.017).

Clinical Validation: Speech alterations occurred independently of ALSFRS-R bulbar subscores.

Results- Vowel Phonation:

Maximum phonation duration (MPT) did not differ between groups, but strong correlations were observed between MPT and FVC% in the ALS group across all vowels:

Conclusion:

Subclinical Sensitivity: Speech metrics capture early bulbar changes before ALSFRS-R bulbar subscore decline.

Respiratory Biomarker: MPT strongly reflect forced vital capacity.

Clinical Trial Utility: Speech metrics offer sensitive endpoints for decentralized ALS trials, enabling continuous longitudinal tracking and early therapeutic signal detection.