Systematic Study of Dysarthric Speech Recognition: Spectral Features and Acoustic Models

2026-06-18

AI Summary

This study addresses the high acoustic variability in dysarthric speech caused by articulatory deviations by systematically evaluating the compatibility of various spectral features with advanced acoustic models. It introduces pitch (fundamental frequency) features into dysarthric speech recognition for the first time and optimizes the frame overlap strategy during training. Employing a Factorized Time Delay Neural Network (F-TDNN) within a hybrid DNN/HMM sequence-discriminative training framework, the proposed approach achieves relative word error rate reductions of 4.65% for isolated words and 4.63% for sentence-level recognition on the TORGO database, significantly mitigating the adverse impact of pronunciation variability on recognition performance.

Abstract

The challenge associated with recognizing dysarthric speech primarily arises from pronounced acoustic variability attributed to impaired articulatory precision. Past research has demonstrated improved recognition through the use of hybrid DNN/HMM sequence discriminative training. This paper presents a comprehensive investigation of various combinations of acoustic features tailored to different Acoustic Models, offering suitable feature selections for each. The incorporation of Pitch features notably improved recognition performance, especially for sentence recognition tasks involving dysarthric speech. Through a systematic examination of the TORGO database, we have demonstrated the potential to enhance the performance of the state-of-the-art Factorized Time Delay Neural Network (F-TDNN) model for recognizing dysarthric speech. Our methods, implemented with the F-TDNN model, resulted in a 4.65\% relative improvement in isolated word recognition and a 4.63\% relative improvement in sentence recognition for dysarthric speech, compared to previous research. This improvement effectively compensates for speech variability, attributable to our deliberate selection of the number of overlapping frames between consecutive training example chunks.

Problem

• dysarthric speech recognition
• acoustic variability
• spectral features
• acoustic models
• speech impairment

Innovation

• dysarthric speech recognition
• spectral features
• F-TDNN
• pitch features
• frame overlap optimization