Detecting of Voice Fatigue With Artificial Intelligence

Area of Science:

• Speech and Hearing Science
• Artificial Intelligence in Medicine
• Machine Learning for Health

Background:

• Voice fatigue (VF) presents diagnostic challenges due to subjective symptoms and difficulty in objective quantification.
• Current methods for detecting VF are limited, necessitating novel approaches for accurate assessment.

Purpose of the Study:

• To develop and evaluate an AI-based system for the automatic detection and monitoring of voice fatigue.
• To compare the performance of the AI model against traditional assessments by speech-language pathologists (SLPs).

Main Methods:

• Collected voice samples from individuals with varying levels of VF.
• Utilized an ECAPA-TDNN model to extract voice embeddings and a Convolutional Neural Network for classification.
• Validated the AI model by comparing its accuracy against assessments from experienced SLPs.

Main Results:

• The AI model achieved 93% accuracy in detecting voice fatigue on a dataset of academic lectures and podcasts.
• The AI model's classification accuracy was 86% when compared to the ratings of three experienced SLPs.

Conclusions:

• AI offers a generalizable and effective approach for the analysis and detection of voice fatigue.
• Future research will focus on validating these AI models with patient data for broader clinical application.