Dysphonia detected by pattern recognition of spectral composition.

Area of Science:

• Speech Acoustics
• Artificial Intelligence
• Bioacoustics

Background:

• Dysphonia affects voice quality and can be challenging to diagnose objectively.
• Traditional acoustic analysis methods may not fully capture the complexity of voice disorders.

Purpose of the Study:

• To investigate the utility of Self-Organizing Maps (SOMs) for visualizing and differentiating normal and dysphonic voices.
• To explore the potential of SOMs in characterizing different types of voice pathologies, such as rough and breathy voices.

Main Methods:

• Utilized Kohonen's Self-Organizing Map algorithm to create two-dimensional acoustic maps from speech data.
• Input consisted of 15-component spectral vectors derived from short-time power spectra at 9.83-msec intervals.
• Analyzed the vowel [a:] from both healthy and dysphonic speakers, examining spectral vector trajectories on the generated maps.

Main Results:

• Dysphonic voices exhibited deviations in spectral composition and stability, visualized as altered trajectory patterns on the SOMs.
• Distinct patterns on the acoustic maps allowed for differentiation between rough and breathy voice qualities.
• While capable of differentiation, the current study could not establish an index for the degree of vocal pathology due to limited speech material.

Conclusions:

• Self-organized acoustic maps offer a powerful on-line visual representation of voice and speech characteristics.
• This visualization method holds promise for diagnostic, educational, and therapeutic applications in voice analysis.
• SOMs provide an easily understandable format for understanding complex voice patterns and potential pathologies.