Energy Band Analysis for Screening Patients Diagnosed with ALS

This paper proposes spectral band analysis of voice signals for the automatic discrimination between patients diagnosed with Amyotrophic Lateral Sclerosis (ALS) and healthy speakers. Tests with repeated cross-validation obtained an average accuracy of 85.2% for a database with all patients diagnosed with Bulbar-type ALS, and with the vowel /a/. This new approach shows results superior to those obtained in the related work where the accuracy has 75%. Using another database with more diversified patients, an average accuracy of 69.4% was obtained with the vowel /a/, and 72.9% with the vowel /i/.