Detection of severe obstructive sleep apnea through voice analysis

Abstract

This paper deals with the potential and limitations of using voice and speech processing to detect Obstructive Sleep Apnea (OSA). An extensive body of voice features has been extracted from patients who present various degrees of OSA as well as healthy controls. We analyze the utility of a reduced set of features for detecting OSA. We apply various feature selection and reduction schemes (statistical ranking, Genetic Algorithms, PCA, LDA) and compare various classifiers (Bayesian Classifiers, kNN, Support Vector Machines, neural networks, Adaboost). S-fold crossvalidation performed on 248 subjects shows that in the extreme cases (that is, 127 controls and 121 patients with severe OSA) voice alone is able to discriminate quite well between the presence and absence of OSA. However, this is not the case with mild OSA and healthy snoring patients where voice seems to play a secondary role. We found that the best classification schemes are achieved using a Genetic Algorithm for feature selection/reduction.

Speech Signal and Facial Image Processing for Obstructive Sleep Apnea Assessment

“…Our results indicate that facial features extracted from frontal and profile images can be better predictors of OSA than acoustic i-vectors features extracted from reading speech. Although different previous studies [ 20 , 21 ], including ours [ 19 ], have reported good results using speech processing techniques for OSA assessment, our recent results, as those reported in this work using a large number of subjects recorded in a clinical practice scenario, only reveal a weak connection between OSA and speech. This fact has also been discussed in our research in [ 41 ] where only very weak correlations were detected between AHI and formant frequencies and bandwidths extracted from sustained vowels.…”

“…Different approaches, generally using similar techniques as in speaker recognition [ 22 ], have been studied for Hebrew [ 16 , 21 ] and Spanish [ 17 ] languages. Results have been reported for different types of speech (i.e., sustained and/or continuous speech) [ 16 , 18 , 20 ], different speech features [ 16 , 19 , 20 ], and modeling different linguistic units [ 18 ]. Also speech recorded from two distinct positions, upright or seated and supine or stretched, has been considered [ 20 , 23 ].…”

Classification techniques on computerized systems to predict and/or to detect Apnea: A systematic review

“…Gaussian) to the histogram of the selected outer beams, and assigning classes to the rest of the beam angles according to the distribution. The observed model is related with a multivariate normal model for the distribution of each class [47] .…”

“…On the other hand, related with ML technologies, the overfitting raised by complex models for the classification of apnea conditions is identified [47] . On the contrary, when inflexible models lead to underfitting is also an observable limitation [13] .…”

Perfil da produção científica da apneia obstrutiva do sono na interface da fonoaudiologia

“…Individuals with OSA present higher risk of cerebrovascular accident 3 (NEUROFUNCTIONAL SPEECH-LANGUAGE PATHOLOGY), and OSA is more frequent with aging 11 (GERONTOLOGY). It is also associated with changes in performance in language levels 12,13 (LANGUAGE), alterations in oropharyngeal muscle tone 14,15 (OROFACIAL MYOLOGY), neurocognitive disorders in children (NEUROPSYCHOLOGY) 16 and differences in acoustic voice parameters 17,18 (VOICE). Considering all these aspects, Speech-Language Pathology should also be involved in training programs for behavioral changes, favoring a better quality of sleep 19 (COMMUNITY HEALTH).…”