Analysis of Articulation Errors in Dysarthric Speech

Area of Science:

• Speech-language pathology
• Acoustic phonetics
• Machine learning in healthcare

Background:

• Imprecise articulation is a primary characteristic of dysarthria, impacting speech intelligibility.
• Accurate detection of articulation errors is crucial for diagnosing and managing dysarthria.
• Acoustic cues from stop consonant production, specifically burst and formant transitions, are key to understanding place of articulation.

Purpose of the Study:

• To propose a novel method for differentiating voiceless stop sounds produced by individuals with dysarthria from those produced by normal speakers.
• To investigate the efficacy of specific acoustic features in identifying dysarthric speech patterns.

Main Methods:

• Extraction of acoustic features, including spectral moments, 2D-DCT of LP spectrum, and Mel frequency cepstral coefficients (MFCCs), from voiceless stops.
• Utilizing a cosine distance-based classifier for speech sample classification.
• Focusing analysis on acoustic deviations around burst and voicing onsets.

Main Results:

• The proposed method successfully differentiates voiceless stops from normal and dysarthric speakers.
• The derived acoustic features effectively capture articulatory errors in dysarthric speech.
• The cosine distance classifier achieved accurate classification of speech samples.

Conclusions:

• The developed feature extraction and classification approach shows promise for automated detection of dysarthric speech.
• This method can serve as a valuable tool for the diagnosis and assessment of articulation disorders in dysarthria.
• Further research can explore broader application of these acoustic features across different speech sound categories and dysarthria subtypes.