Artificial Dysarthric Speech Recognition using Acoustic Descriptors and Neural Networks

Aguero-Terrones S.; Paranjape R. · Canadian Conference on Electrical and Computer Engineering (2026) · DOI: 10.1109/ccece68150.2026.11610140

Resumen

Currently, voice-based interfaces are increasingly used in applications that seek to improve accessibility, particularly in the development of assistive technologies for people with speech difficulties. Regarding that, the present work proposes a speech recognition system for words focused on dysarthric speech, a speech disorder that commonly appears in people with conditions and diseases such as cerebral palsy (CP) and amyotrophic lateral sclerosis (ALS). Furthermore, a speech database was constructed using a transformation method to obtain artificial dysarthric speech. Consequently, the speech energy of the transformed speech was calculated to detect the beginning and end of each word to proceed with acoustic descriptor extraction and selection of the most useful descriptors were carried out. Finally, a neural network was trained to classify the words into four different voice commands. The proposed system represents an alternative for dysarthric speech recognition in voice-based assistive systems.

Instituciones: Universidad Peruana de Ciencias Aplicadas