Nonlinear dynamic modeling of impaired voice

Area of Science:

• Speech Science
• Biomedical Engineering
• Nonlinear Dynamics

Background:

• Patients with voice control muscle issues often exhibit speech pathologies.
• Existing linear speech production models do not fully capture complex vocal dynamics.

Purpose of the Study:

• To develop a nonlinear dynamic model for modeling vowels from patients with laryngeal muscle control problems.
• To enhance automatic speech recognition (ASR) and speech pathology detection systems.
• To improve communication capabilities for patients with speech impairments.

Main Methods:

• A nonlinear dynamic model incorporating feedback and a sigmoid nonlinear function was developed.
• The model was trained and applied to discriminate between normal and pathological voice samples.
• Statistical analysis of model parameters was performed to assess feature distinctiveness.

Main Results:

• The nonlinear model successfully captured higher-order dynamics of the speech signal.
• Feedback integration enhanced the model's dynamicity.
• Model parameters provided independent and distinct features for discriminating pathological voice classes.

Conclusions:

• The proposed nonlinear dynamic model is effective for analyzing pathological speech.
• The model's parameters offer valuable features for speech pathology detection and ASR.
• This approach has the potential to significantly improve assistive communication technologies.