Automatic speech and singing classification in ambulatory recordings for normal and disordered voices

Area of Science:

• Otolaryngology
• Speech and Hearing Sciences
• Biomedical Engineering

Background:

• Phonotraumatic vocal hyperfunction (PVH) is linked to vocal fold lesions, particularly in professional singers.
• Ambulatory voice monitoring offers a non-invasive method for studying vocal behavior in real-world settings.
• Understanding the phonatory mechanisms in PVH is crucial for effective diagnosis and treatment.

Purpose of the Study:

• To develop and validate a classifier for identifying phonotraumatic vocal hyperfunction (PVH) in singers using ambulatory voice monitoring.
• To differentiate between singers with and without PVH based on acoustic voice features.
• To assess the utility of neck-surface accelerometer data for diagnosing vocal hyperfunction.

Main Methods:

• Twenty singers with PVH and 20 matched healthy controls were enrolled.
• Ambulatory voice data were collected using a neck-surface accelerometer-based voice monitor.
• A logistic regression classifier was trained using fundamental frequency and autocorrelation peak amplitude from an expert-labeled dataset.

Main Results:

• The developed classifier achieved an overall accuracy of 94.2% on a held-out test set.
• Key input features for classification included fundamental frequency and autocorrelation peak amplitude.
• The study demonstrated the feasibility of distinguishing PVH in singers via ambulatory monitoring.

Conclusions:

• Ambulatory voice monitoring, coupled with a trained classifier, is a highly accurate tool for detecting phonotraumatic vocal hyperfunction in singers.
• This technology can provide valuable insights into the phonatory characteristics associated with vocal fold lesions.
• The findings support the use of objective voice monitoring in the clinical assessment and management of professional vocalists with PVH.