Obstructive sleep apnea severity estimation: Fusion of speech-based systems

Area of Science:

• Sleep Medicine
• Acoustic Analysis
• Biomedical Engineering

Background:

• Obstructive sleep apnea (OSA) is a prevalent sleep disorder linked to upper airway anatomical issues.
• These anatomical variations may manifest as detectable changes in speech acoustics.
• Early OSA assessment is crucial for effective management.

Purpose of the Study:

• To investigate if speech signals contain information for estimating the apnea-hypopnea index (AHI).
• To develop a noninvasive method for assessing OSA severity using speech analysis.
• To evaluate the diagnostic agreement between speech-estimated AHI and polysomnography (PSG)-determined AHI.

Main Methods:

• 198 male participants referred for polysomnography (PSG) provided speech recordings.
• Speech samples (vowels, fluent speech segments, full recording) were analyzed for acoustic features.
• Support vector regression and regression trees were employed to estimate AHI from speech features.

Main Results:

• A diagnostic agreement of 67.3% was achieved between speech-estimated AHI and PSG-determined AHI.
• The developed method demonstrated an absolute error rate of 10.8 events/hr.
• Fusion of features from different speech segments improved AHI estimation.

Conclusions:

• Speech signal analysis holds promise for assisting in AHI estimation.
• This approach could lead to the development of a noninvasive screening tool for OSA.
• Further research can refine speech analysis techniques for more accurate OSA assessment.