Title: Multilingual voice disorder classification using glottal flow and MFCC-based acoustic analysis

Authors: Nitin Pal; Girish Gidaye; Varsha Turkar; Uma Jaishankar

Addresses: Department of Electronics and Computer Science, Vidyalankar Institute of Technology, Mumbai, India ' Department of Electronics and Computer Science, Vidyalankar Institute of Technology, Mumbai, India ' Electronics and Telecommunication Engineering Department, Vidyalankar Institute of Technology, Mumbai, India ' Department of Electronics Engineering, Vidyalankar Institute of Technology, Mumbai, India

Abstract: Vocal pathologies affect vocal fold dynamics, altering pitch, loudness, and voice quality. Conventional methods rely on invasive techniques. Many researchers have used machine learning models based on features extracted from speech signals. It may not fully capture physiological alterations in vocal fold impairments. To address these challenges, the work in this paper evaluates glottal flow features mined from true voice sources by comparing them against Mel-frequency cepstral coefficients (MFCC) based features across four linguistically diverse datasets. The proposed non-invasive method captures most physiological alterations in vocal fold impairments as the features are derived from true voice sources. The data augmentation, oversampling techniques and min-max normalisation are employed to overcome dataset limitations and improve model generalisation. Sustained vowel /a/ samples are used to train multiple classifiers for each dataset for comparative analysis. It is observed that classifiers using glottal flow features achieved superior performance compared to MFCC.

Keywords: vocal pathology; classification; glottal flow features; pathological speech analysis; LSTM model; voice disorder detection; healthcare AI applications.

DOI: 10.1504/IJBET.2025.149595

International Journal of Biomedical Engineering and Technology, 2025 Vol.49 No.2, pp.95 - 120

Received: 02 Mar 2025
Accepted: 16 May 2025
Published online: 07 Nov 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article