Uncovering Voice Misuse Using Symbolic Mismatch

Area of Science:

• Biomedical Engineering
• Speech Science
• Data Mining

Background:

• Voice disorders impact millions globally, affecting working-aged individuals.
• Current diagnostic methods for behavioral voice disorders can be subjective.
• Objective, data-driven approaches are needed for diagnosis and treatment evaluation.

Purpose of the Study:

• To present the first large-scale study of vocal misuse using long-term ambulatory accelerometer data.
• To investigate an unsupervised data mining approach for uncovering latent information about voice misuse.
• To establish an objective basis for diagnosing behavioral voice disorders and understanding voice therapy.

Main Methods:

• Collected over 253 days of ambulatory data from 22 subjects using neck-worn accelerometers.
• Segmented signals into over a hundred million single glottal pulses (vocal fold closures).
• Employed unsupervised clustering and symbolic mismatch to compare patients and controls, and pre- vs. post-treatment.

Main Results:

• Identified significant behavioral differences between patients with voice disorders and matched controls.
• Observed significant differences in some patients before and after treatment.
• Demonstrated the potential of symbolic mismatch analysis for differentiating vocal behaviors.

Conclusions:

• The unsupervised data mining approach offers an objective method for diagnosing behavioral voice disorders.
• This study is a foundational step towards a data-driven understanding of voice therapy's impact.
• Ambulatory accelerometer data provides valuable insights into vocal misuse patterns.