Can Machine Learning Assist Locating the Excitation of Snore Sound? A Review.

In the past three decades, snoring (affecting more than 30 % adults of the UK population) has been increasingly studied in the transdisciplinary research community involving medicine and engineering. Early work demonstrated that, the snore sound can carry important information about the status of the upper airway, which facilitates the development of non-invasive acoustic based approaches for diagnosing and screening of obstructive sleep apnoea and other sleep disorders. Nonetheless, there are more demands from clinical practice on finding methods to localise the snore sound's excitation rather than only detecting sleep disorders. In order to further the relevant studies and attract more attention, we provide a comprehensive review on the state-of-the-art techniques from machine learning to automatically classify snore sounds. First, we introduce the background and definition of the problem. Second, we illustrate the current work in detail and explain potential applications. Finally, we discuss the limitations and challenges in the snore sound classification task. Overall, our review provides a comprehensive guidance for researchers to contribute to this area.

Snore-GANs: Improving Automatic Snore Sound Classification With Synthesized Data.

One of the frontier issues that severely hamper the development of automatic snore sound classification (ASSC) associates to the lack of sufficient supervised training data. To cope with this problem, we propose a novel data augmentation approach based on semi-supervised conditional generative adversarial networks (scGANs), which aims to automatically learn a mapping strategy from a random noise space to original data distribution. The proposed approach has the capability of well synthesizing "realistic" high-dimensional data, while requiring no additional annotation process. To handle the mode collapse problem of GANs, we further introduce an ensemble strategy to enhance the diversity of the generated data. The systematic experiments conducted on a widely used Munich-Passau snore sound corpus demonstrate that the scGANs-based systems can remarkably outperform other classic data augmentation systems, and are also competitive to other recently reported systems for ASSC.

A Bag of Wavelet Features for Snore Sound Classification.

Snore sound (SnS) classification can support a targeted surgical approach to sleep related breathing disorders. Using machine listening methods, we aim to find the location of obstruction and vibration within a subject's upper airway. Wavelet features have been demonstrated to be efficient in the recognition of SnSs in previous studies. In this work, we use a bag-of-audio-words approach to enhance the low-level wavelet features extracted from SnS data. A Naïve Bayes model was selected as the classifier based on its superiority in initial experiments. We use SnS data collected from 219 independent subjects under drug-induced sleep endoscopy performed at three medical centres. The unweighted average recall achieved by our proposed method is 69.4%, which significantly ([Formula: see text] one-tailed z-test) outperforms the official baseline (58.5%), and beats the winner (64.2%) of the INTERSPEECH COMPARE Challenge 2017 Snoring sub-challenge. In addition, the conventionally used features like formants, mel-scale frequency cepstral coefficients, subband energy ratios, spectral frequency features, and the features extracted by the OPENSMILE toolkit are compared with our proposed feature set. The experimental results demonstrate the effectiveness of the proposed method in SnS classification.

Snoring - An Acoustic Definition.

Objective- The distinction of snoring and loud breathing is often subjective and lies in the ear of the beholder. The aim of this study is to identify and assess acoustic features with a high suitability to distinguish these two classes of sound, in order to facilitate an objective definition of snoring based on acoustic parameters. Methods- A corpus of snore and breath sounds from 23 subjects has been used that were classified by 25 human raters. Using the openSMILE feature extractor, 6 373 acoustic features have been evaluated for their selectivity comparing SVM classification, logistic regression, and the recall of each single feature. Results- Most selective single features were several statistical functionals of the first and second mel frequency spectrum-generated perceptual linear predictive (PLP) cepstral coefficient with an unweighted average recall (UAR) of up to 93.8%. The best performing feature sets were low level descriptors (LLDs), derivatives and statistical functionals based on fast Fourier transformation (FFT), with a UAR of 93.0%, and on the summed mel frequency spectrum-generated PLP cepstral coefficients, with a UAR of 92.2% using SVM classification. Compared to SVM classification, logistic regression did not show considerable differences in classification performance. Conclusion- It could be shown that snoring and loud breathing can be distinguished by robust acoustic features. The findings might serve as a guidance to find a consensus for an objective definition of snoring compared to loud breathing.

Snoring classified: The Munich-Passau Snore Sound Corpus.

OBJECTIVE: Snoring can be excited in different locations within the upper airways during sleep. It was hypothesised that the excitation locations are correlated with distinct acoustic characteristics of the snoring noise. To verify this hypothesis, a database of snore sounds is developed, labelled with the location of sound excitation. METHODS: Video and audio recordings taken during drug induced sleep endoscopy (DISE) examinations from three medical centres have been semi-automatically screened for snore events, which subsequently have been classified by ENT experts into four classes based on the VOTE classification. The resulting dataset containing 828 snore events from 219 subjects has been split into Train, Development, and Test sets. An SVM classifier has been trained using low level descriptors (LLDs) related to energy, spectral features, mel frequency cepstral coefficients (MFCC), formants, voicing, harmonic-to-noise ratio (HNR), spectral harmonicity, pitch, and microprosodic features. RESULTS: An unweighted average recall (UAR) of 55.8% could be achieved using the full set of LLDs including formants. Best performing subset is the MFCC-related set of LLDs. A strong difference in performance could be observed between the permutations of train, development, and test partition, which may be caused by the relatively low number of subjects included in the smaller classes of the strongly unbalanced data set. CONCLUSION: A database of snoring sounds is presented which are classified according to their sound excitation location based on objective criteria and verifiable video material. With the database, it could be demonstrated that machine classifiers can distinguish different excitation location of snoring sounds in the upper airway based on acoustic parameters.

Snore sound recognition: On wavelets and classifiers from deep nets to kernels.

In this paper, we present a comprehensive comparison of wavelet features for the classification of snore sounds. Wavelet features have proven to be efficient in our previous work; however, the benefits of wavelet transform energy (WTE) and wavelet packet transform energy (WPTE) features were not clearly established. In this study, we firstly present our updated snore sounds database, expanded from 24 patients (collected by one medical centre) to 40 patients (collected by three medical centres). We then study the effects of varying frame sizes and overlaps for extraction of the wavelet low-level descriptors, the effect of which have yet to be fully established. We also compare the performance of the WTE and WPTE features when fed into multiple classifiers, namely, Support Vector Machines (SVM), K-Nearest Neighbours, Linear Discriminant Analysis, Random Forests, Extreme Learning Machines, Kernel Extreme Learning Machines, Multilayer Perceptron, and Deep Neural Networks. Key results presented indicate that, when fed into a SVM, WTE outperforms WPTE (one-tailed z-test, p<;0.002). Further, WPTE can achieve a significant improvement when trained by a k-nearest neighbours classifier (one-tailed z-test, p <; 0.001).

Classification of the Excitation Location of Snore Sounds in the Upper Airway by Acoustic Multifeature Analysis.

OBJECTIVE: Obstructive sleep apnea (OSA) is a serious chronic disease and a risk factor for cardiovascular diseases. Snoring is a typical symptom of OSA patients. Knowledge of the origin of obstruction and vibration within the upper airways is essential for a targeted surgical approach. Aim of this paper is to systematically compare different acoustic features, and classifiers for their performance in the classification of the excitation location of snore sounds. METHODS: Snore sounds from 40 male patients have been recorded during drug-induced sleep endoscopy, and categorized by Ear, Nose & Throat (ENT) experts. Crest Factor, fundamental frequency, spectral frequency features, subband energy ratio, mel-scale frequency cepstral coefficients, empirical mode decomposition-based features, and wavelet energy features have been extracted and fed into several classifiers. Using the ReliefF algorithm, features have been ranked and the selected feature subsets have been tested with the same classifiers. RESULTS: A fusion of all features after a ReliefF feature selection step in combination with a random forests classifier showed the best classification results of 78% unweighted average recall by subject independent validation. CONCLUSION: Multifeature analysis is a promising means to help identify the anatomical mechanisms of snore sound generation in individual subjects. SIGNIFICANCE: This paper describes a novel approach for the machine-based multifeature classification of the excitation location of snore sounds in the upper airway.