New advances on detecting obstructive sleep apnea based on acoustic information

doi:10.11958/20250966

Abstract

Abstract:

Obstructive sleep apnea (OSA) is a common sleep disorder characterized by repeated episodes of upper airway collapse and obstruction during sleep. Polysomnography is the gold standard for diagnosing OSA, but it is expensive, time-consuming, and can cause discomfort for patients. In recent years, acoustic-based approaches for detecting OSA have emerged as a research focus. This review summarizes recent advances in OSA automatic detection techniques based on snoring and speech signals, and systematically examines their applications in diagnosis, severity assessment, and localization of obstruction sites. Findings indicate that the acoustic features of snoring and speech signals hold significant value for OSA screening, and when combined with machine learning and deep learning models, it can achieve high diagnostic accuracy. Future research should focus on elucidating the relationship between acoustic features and the pathophysiological mechanisms of OSA, integrating multimodal information, and advancing the clinical application of wearable devices, with the aim of promoting intelligent, non-invasive, and cost-effective screening technologies for OSA.

Key words: sleep apnea, obstructive, snoring, speech, detection, obstruction location, severity

CLC Number:

R563.8

Figures/Tables 5

References 52

[1]	FRANKLIN K A, LINDBERG E. Obstructive sleep apnea is a common disorder in the population-a review on the epidemiology of sleep apnea[J]. J Thorac Dis, 2015, 7(8):1311-1322. doi:10.3978/j.issn.2072-1439.2015.06.11.
[2]	YEGHIAZARIANS Y, JNEID H, TIETJENS J R, et al. Obstructive sleep apnea and cardiovascular disease: a scientific statement from the American Heart Association[J]. Circulation, 2021, 144(3):e56-e67. doi:10.1161/CIR.0000000000000988.
[3]	PEVERNAGIE D, AARTS R M, DE MEYER M. The acoustics of snoring[J]. Sleep Med Rev, 2010, 14(2):131-144. doi:10.1016/j.smrv.2009.06.002.
[4]	KAPUR V K, AUCKLEY D H, CHOWDHURI S, et al. Clinical practice guideline for diagnostic testing for adult obstructive sleep apnea:an american academy of sleep medicine clinical practice guideline[J]. J Clin Sleep Med, 2017, 13(3):479-504. doi:10.5664/jcsm.6506.
[5]	CHIANG J K, LIN Y C, LU C M, et al. Correlation between snoring sounds and obstructive sleep apnea in adults:a meta-regression analysis[J]. Sleep Sci, 2022, 15(4):463-470. doi:10.5935/1984-0063.20220068.
[6]	GÜRPıNAR B, SALTÜRK Z, KUMRAL T L, et al. Analysis of snoring to determine the site of obstruction in obstructive sleep apnea syndrome[J]. Sleep Breath, 2021, 25(3):1427-1432. doi:10.1007/s11325-020-02252-5.
[7]	PEREZ-PADILLA J R, SLAWINSKI E, DIFRANCESCO L M, et al. Characteristics of the snoring noise in patients with and without occlusive sleep apnea[J]. Am Rev Respir Dis, 1993, 147(3):635-644. doi:10.1164/ajrccm/147.3.635.
[8]	FIZ J A, ABAD J, JANÉ R, et al. Acoustic analysis of snoring sound in patients with simple snoring and obstructive sleep apnoea[J]. Eur Respir J, 1996, 9(11):2365-2370. doi:10.1183/09031936.96.09112365.
[9]	NG A K, KOH T S, BAEY E, et al. Could formant frequencies of snore signals be an alternative means for the diagnosis of obstructive sleep apnea?[J]. Sleep Med, 2008, 9(8):894-898. doi:10.1016/j.sleep.2007.07.010.
[10]	HERZOG M, SCHMIDT A, BREMERT T, et al. Analysed snoring sounds correlate to obstructive sleep disordered breathing[J]. Eur Arch Otorhinolaryngol, 2008, 265(1):105-113. doi:10.1007/s00405-007-0408-8.
[11]	DING L, PENG J, SONG L, et al. Automatically detecting OSAHS patients based on transfer learning and model fusion[J]. Physiol Meas, 2024, 45(5):055013. doi:10.1088/1361-6579/ad4953.
[12]	DING L, PENG J, SONG L, et al. Automatically detecting apnea-hypopnea snoring signal based on VGG19 + LSTM[J]. Biomed Signal Process Control, 2023,80:104351. doi:10.1016/j.bspc.2022.104351.
[13]	SONG Y, SUN X, DING L, et al. AHI estimation of OSAHS patients based on snoring classification and fusion model[J]. Am J Otolaryngol, 2023, 44(5):103964. doi:10.1016/j.amjoto.2023.103964.
[14]	LUO H W, LI H, LU Y, et al. Design of embedded real-time system for snoring and OSA detection based on machine learning[J]. Measurement, 2023,214:112802. doi:10.1016/j.measure-ment.2023.112802.
[15]	SUN X, DING L, SONG Y, et al. Automatic identifying OSAHS patients and simple snorers based on Gaussian mixture models[J]. Physiol Meas, 2023, 44(4):045003. doi:10.1088/1361-6579/accd43.
[16]	CHENG S, WANG C, YUE K, et al. Automated sleep apnea detection in snoring signal using long short-term memory neural networks[J]. Biomedical Signal Processing and Control, 2022, 71(Jan. Pt. B):103238.1-103238.9. doi:10.1016/j.bspc.2021.103238.
[17]	CASTILLO-ESCARIO Y, WERTHEN-BRABANTS L, GROENENDAAL W, et al. Convolutional neural networks for apnea detection from smartphone audio signals: effect of window size[C/OL]//2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC). 2022:666-669[2024-11-28]. https://ieeexplore.ieee.org/document/9871396.
[18]	JIANG Y, PENG J, SONG L. An OSAHS evaluation method based on multi-features acoustic analysis of snoring sounds[J]. Sleep Med, 2021, 84:317-323. doi:10.1016/j.sleep.2021.06.012.
[19]	SHEN F, CHENG S, LI Z, et al. Detection of snore from OSAHS patients based on deep learning[J]. J Healthc Eng, 2020,2020:8864863. doi:10.1155/2020/8864863.
[20]	JIANG Y, PENG J, ZHANG X. Automatic snoring sounds detection from sleep sounds based on deep learning[J]. Phys Eng Sci Med, 2020, 43(2):679-689. doi:10.1007/s13246-020-00876-1.
[21]	BAHR-HAMM K, ABRIANI A, ANWAR A R, et al. Using entropy of snoring, respiratory effort and electrocardiography signals during sleep for OSA detection and severity classification[J]. Sleep Med, 2023, 111:21-27. doi:10.1016/j.sleep.2023.09.005.
[22]	YE Z, PENG J, ZHANG X, et al. Identification of OSAHS patients based on ReliefF-mRMR feature selection[J]. Phys Eng Sci Med, 2024, 47(1):99-108. doi:10.1007/s13246-023-01345-1.
[23]	LUO J, LIU H, GAO X, et al. A novel deep feature transfer-based OSA detection method using sleep sound signals[J]. Physiol Meas, 2020, 41(7):075009. doi:10.1088/1361-6579/ab9e7b.
[24]	CHO S W, JUNG S J, SHIN J H, et al. Evaluating prediction models of sleep apnea from smartphone-recorded sleep breathing sounds[J]. JAMA Otolaryngol Head Neck Surg, 2022, 148(6):515-521. doi:10.1001/jamaoto.2022.0244.
[25]	HOU L, PAN Q, YI H, et al. Estimating a sleep apnea hypopnea index based on the erb correlation dimension of snore sounds[J]. Front Digit Health, 2020,2:613725. doi:10.3389/fdgth.2020.613725.
[26]	QIU X, WANG C, LI B, et al. An audio-semantic multimodal model for automatic obstructive sleep Apnea-Hypopnea Syndrome classification via multi-feature analysis of snoring sounds[J]. Front Neurosci, 2024,18:1336307. doi:10.3389/fnins.2024.1336307.
[27]	FANG L, CAI J, HUANG Z, et al. Assessment of simulated snoring sounds with artificial intelligence for the diagnosis of obstructive sleep apnea[J]. Sleep Med, 2025, 125:100-107. doi:10.1016/j.sleep.2024.11.018.
[28]	LE V L, KIM D, CHO E, et al. Real-time detection of sleep apnea based on breathing sounds and prediction reinforcement using home noises: algorithm development and validation[J]. J Med Internet Res, 2023,25:e44818. doi:10.2196/44818.
[29]	ROMERO H E, MA N, BROWN G J, et al. Acoustic screening for obstructive sleep apnea in home environments based on deep neural networks[J]. IEEE J Biomed Health Inform, 2022, 26(7):2941-2950. doi:10.1109/JBHI.2022.3154719.
[30]	WANG B, TANG X, AI H, et al. Obstructive sleep apnea detection based on sleep sounds via deep learning[J]. Nat Sci Sleep, 2022, 14:2033-2045. doi:10.2147/NSS.S373367.
[31]	LIU Y, FENG Y, LI Y, et al. Automatic classification of the obstruction site in obstructive sleep apnea based on snoring sounds[J]. Am J Otolaryngol, 2022, 43(6):103584. doi:10.1016/j.amjoto.2022.103584.
[32]	DING L, PENG J X. Automatic classification of snoring sounds from excitation locations based on prototypical network[J]. Applied Acoustics, 2022,195:108799. doi:10.1016/j.apacoust.2022.108799.
[33]	SEBASTIAN A, CISTULLI P A, COHEN G, et al. Automated identification of the predominant site of upper airway collapse in obstructive sleep apnoea patients using snore signal[J]. Physiol Meas, 2020, 41(9):095005. doi:10.1088/1361-6579/abaa33.
[34]	SUN J, HU X, CHEN C, et al. Amplitude spectrum trend-based feature for excitation location classification from snore sounds[J]. Physiol Meas, 2020, 41(8):085006. doi:10.1088/1361-6579/abaa34.
[35]	SUN J, HU X, PENG S, et al. Automatic classification of excitation location of snoring sounds[J]. J Clin Sleep Med, 2021, 17(5):1031-1038. doi:10.5664/jcsm.9094.
[36]	FOX A W, MONOSON P K, MORGAN C D. Speech dysfunction of obstructive sleep apnea. A discriminant analysis of its descriptors[J]. Chest, 1989, 96(3):589-595. doi:10.1378/chest.96.3.589.
[37]	GOLDSHTEIN E, TARASIUK A, ZIGEL Y. Automatic detection of obstructive sleep apnea using speech signals[J]. IEEE Trans Biomed Eng, 2011, 58(5):1373-1382. doi:10.1109/TBME.2010.2100096.
[38]	MONOSON P K, FOX A W. Preliminary observation of speech disorder in obstructive and mixed sleep apnea[J]. Chest, 1987, 92(4):670-675. doi:10.1378/chest.92.4.670.
[39]	FIZ J A, MORERA J, ABAD J, et al. Acoustic analysis of vowel emission in obstructive sleep apnea[J]. Chest, 1993, 104(4):1093-1096. doi:10.1378/chest.104.4.1093.
[40]	POZO R F, MURILLO J L B, GOMEZ L H, et al. Assessment of severe apnoea through voice analysis,automatic speech, and speaker recognition techniques[J]. EURASIP Journal on Advances in Signal Processing, 2009, 2009(10):982531:1-982531:11.
[41]	DING Y, WANG J, GAO J, et al. Severity evaluation of obstructive sleep apnea based on speech features[J]. Sleep Breath, 2021, 25(2):787-795. doi:10.1007/s11325-020-02168-0.
[42]	UĞUR T K, YILMAZ D, YILDIZ M, et al. A preliminary study on OSA severity levels detection by evaluating speech signals nonlinearities with multi-class classification[J]. IEEE Access, 2023, 11:120997-121012. doi:10.1109/ACCESS.2023.3327902.
[43]	YILMAZ D, YILDIZ M, UYAR TOPRAK Y, et al. Obstructive sleep apnea detection with nonlinear analysis of speech[J]. Biomedical Signal Processing and Control, 2023,84:104956. doi:10.1016/j.bspc.2023.104956.
[44]	ZHANG K, CAO L, DING Y, et al. Obstructive sleep apnea detection using pre-trained speech representations[J]. 2023, 2023(8):1139-1143. doi:10.21437/Interspeech.2023-278.
[45]	BOTELHO C, ABAD A, SCHULTZ T, et al. Visual speech for obstructive sleep apnea detection[J]. Interspeech, 2021, 2021:2516-2520. doi:10.21437/Interspeech.2021-1717.
[46]	PANG K G, HSUNG T C, LAW A K W, et al. Optimal vowels measurements for obstructive sleep apnea detection using speech signals[J]. 2020 IEEE 3rd International Conference on Information Communication and Signal Processing (ICICSP), 2020,2020:143-147. doi:10.1109/1CICSP50920.2020.9231972.
[47]	SIMPLY R M, DAFNA E, ZIGEL Y. Diagnosis of obstructive sleep apnea using speech signals from awake subjects[J]. IEEE Journal of Selected Topics in Signal Processing, 2020, 14(2):251-260. doi:10.1109/JSTSP2019.2955019.
[48]	PERERO-CODOSERO J M, ESPINOZA-CUADROS F, ANTON-MARTIN J, et al. Modeling obstructive sleep apnea voices using deep neural network embeddings and domain-adversarial training[J]. IEEE Journal of Selected Topics in Signal Processing, 2020, 14(2):240-250. doi:10.1109/JSTSP.2019.2957977.
[49]	ESPINOZA-CUADROS F, FERNÁNDEZ-POZO R, TOLEDANO D T, et al. Reviewing the connection between speech and obstructive sleep apnea[J]. Biomed Eng Online, 2016,15:20. doi:10.1186/s12938-016-0138-5.
[50]	BEN OR D, DAFNA E, TARASIUK A, et al. Obstructive sleep apnea severity estimation: Fusion of speech-based systems[J]. Annu Int Conf IEEE Eng Med Biol Soc, 2016, 2016:3207-3210. doi:10.1109/EMBC.2016.7591411.
[51]	SOLE-CASALS J, MUNTEANU C, MARTIN O C, et al. Detection of severe obstructive sleep apnea through voice analysis[J]. Applied Soft Computing, 2014, 23:346-354. doi:10.1016/j.asoc.2014.06.017.
[52]	CAO S, ROSENZWEIG I, BILOTTA F, et al. Automatic detection of obstructive sleep apnea based on speech or snoring sounds: a narrative review[J]. J Thorac Dis, 2024, 16(4):2654-2667. doi:10.21037/jtd-24-310.

作者	样本分布	特征	模型	模型识别OSA性能
Ding等^[11]	正常7例，轻度OSA 6例，中度17例，重度65例	MFCC，VGG16和PANN提取的深度特征	特征筛选：XGBoost；分类：KNN，RF	MFCC-KNN、VGG16-KNN和PANN-KNN识别准确度为100%
Ding等^[12]	正常10例，轻度OSA 6例，中度10例，重度24例	梅尔频谱图	VGG19+LSTM	识别准确度99.31%，敏感度99.13%，特异度99.58%
Song等^[13]	受试者40例， AHI为37.9±28.9	MFCC，PLP，BSF，PR800，F₀等	3个模型融合：声学特征+ XGBoost，梅尔频谱图+CNN，梅尔频谱图+ ResNet18	准确度83.44%，召回率85.27%
Luo等^[14]	受试者32例，AHI为32.8±24.4	MFCC	TCN	OSA相关打鼾检测准确度96.7%
Sun等^[15]	正常6例，OSA 24例	MFCC，PLP，BSF SE，PR800，F₀，共振峰 GTCC	特征筛选：Fisher比率；分类：GMM	OSA识别准确度90.0%，精确度95.7%
Cheng等^[16]	正常10例，OSA 33例	MFCC，Fbanks，LPC	LSTM	呼吸事件相关打鼾和正常打鼾的分类准确度为95.3%
Castillo-Escario等^[17]	正常2例，轻度OSA 3例，中度13例，重度7例	频谱图	CNN	OSA识别的准确度、敏感度和特异度分别为88.5%、71.8%和89.1%。
Jiang等^[18]	正常4例，OSA 8例	LPC，PR800，MFCC，SE	特征筛选：随机森林分类：LR、SVM、贝叶斯、KNN、ANN	LR的分类准确度、敏感度、特异度均达到100%
Shen等^[19]	正常16例，OSA 16例	MFCC	LSTM	分类准确度、敏感度和特异度分别为87%、84%和91%
Jiang等^[20]	正常4例，OSA 11例	时域波形，频谱、声谱、梅尔频谱及CQT声谱图	CNNs-DNNs，CNNs-LSTMs-DNNs	梅尔频谱图结合CNNs-LSTMs-DNNs效果最好，准确度达到95.07%

作者	样本分布	特征	模型	模型识别OSA性能
Ding等^[11]	正常7例，轻度OSA 6例，中度17例，重度65例	MFCC，VGG16和PANN提取的深度特征	特征筛选：XGBoost；分类：KNN，RF	MFCC-KNN、VGG16-KNN和PANN-KNN识别准确度为100%
Ding等^[12]	正常10例，轻度OSA 6例，中度10例，重度24例	梅尔频谱图	VGG19+LSTM	识别准确度99.31%，敏感度99.13%，特异度99.58%
Song等^[13]	受试者40例， AHI为37.9±28.9	MFCC，PLP，BSF，PR800，F₀等	3个模型融合：声学特征+ XGBoost，梅尔频谱图+CNN，梅尔频谱图+ ResNet18	准确度83.44%，召回率85.27%
Luo等^[14]	受试者32例，AHI为32.8±24.4	MFCC	TCN	OSA相关打鼾检测准确度96.7%
Sun等^[15]	正常6例，OSA 24例	MFCC，PLP，BSF SE，PR800，F₀，共振峰 GTCC	特征筛选：Fisher比率；分类：GMM	OSA识别准确度90.0%，精确度95.7%
Cheng等^[16]	正常10例，OSA 33例	MFCC，Fbanks，LPC	LSTM	呼吸事件相关打鼾和正常打鼾的分类准确度为95.3%
Castillo-Escario等^[17]	正常2例，轻度OSA 3例，中度13例，重度7例	频谱图	CNN	OSA识别的准确度、敏感度和特异度分别为88.5%、71.8%和89.1%。
Jiang等^[18]	正常4例，OSA 8例	LPC，PR800，MFCC，SE	特征筛选：随机森林分类：LR、SVM、贝叶斯、KNN、ANN	LR的分类准确度、敏感度、特异度均达到100%
Shen等^[19]	正常16例，OSA 16例	MFCC	LSTM	分类准确度、敏感度和特异度分别为87%、84%和91%
Jiang等^[20]	正常4例，OSA 11例	时域波形，频谱、声谱、梅尔频谱及CQT声谱图	CNNs-DNNs，CNNs-LSTMs-DNNs	梅尔频谱图结合CNNs-LSTMs-DNNs效果最好，准确度达到95.07%

作者	发表年份	样本分布	特征	模型	模型性能
Bahr-Hamm等^[21]	2023	正常（R）15例，轻度OSA（L） 27例，中度（M）21例，重度（H）23例	鼾声、心电图和胸腹部偏移信号的采样熵	SVM	AUC：R vs. L：61%，R vs. M：68%，R vs.H：84%，L vs. M：63%，L vs. H：82%，M vs. H：65%
YE等^[22]	2024	正常9例，轻度OSA 11例，中度19例，重度55例	MFCC，LPC，SE， PR800，F₀，共振峰等	ReliefF-mRMR 特征筛选，SVM，LR，KNN，NB分类	SVM在AHI分别为5、15、30的阈值下，识别OSA准确度分别为100%、100%和95.94%
Luo等^[23]	2020	受试者132例，其中男109例，AHI为40.62±26.62，女23例，AHI为24.73±27.64	CNN提取的深度特征	逻辑回归	在AHI分别为c15、30的阈值下，分类准确度分别为80.17%和80.21%
Cho等^[24]	2022	受试者423例，AHI为32.6±24.4	LPC，MFCC，SE等	随机森林	在AHI分别为5、15、30的阈值下，识别OSA准确度分别为88.2%、82.3%和81.7%
Hou等^[25]	2021	正常10例，轻度OSA 23例，中度24例，重度36例	ECD	GMM	分类准确度达87.74%，其中正常、轻度、中度和重度的准确度分别为100%，81.48%，75.75%和93.75%
Qiu等^[26]	2024	正常26例，轻度OSA 45例，中度44例，重度82例	经PubMedBERT 转换的MFCC	XGBoost	分类准确度为58%，正常、轻度、中度和重度的准确度分别为50%、50%、66.7%和75%
Fang等^[27]	2025	正常40例，轻度OSA 40例，中度72例，重度313例	梅尔频谱图	AST	在AHI分别为5、15、30的阈值下，识别OSA准确度分别为92.6%、88.7%和83.0%
Le等^[28]	2023	正常282例，轻度OSA 316例，中度316例，重度401例	梅尔频谱图	DNN+线性回归模型	按AHI分别为5、15和30划分，敏感度为0.97、0.85、0.96，特异度为0.89、0.84、0.91
Romero等^[29]	2022	正常13例，轻度OSA 67例，中度38例，重度39例	梅尔滤波器组特征	DNN	AHI=15：敏感度79%，特异度80%； AHI=30：敏感度78%，特异度93%
Wang等^[30]	2022	训练组116例，AHI为27.6（9.9~61.6）；验证组19例，AHI为18.9（5.9~62.1）；测试组59例，AHI为24.4（6.6~54.3）	梅尔频谱图	OSAnet	在AHI=5、10，15、30的阈值下，识别OSA敏感度分别为93.6%、82.5%、88.5%和95.6%

作者	发表年份	样本分布	特征	模型	模型性能
Bahr-Hamm等^[21]	2023	正常（R）15例，轻度OSA（L） 27例，中度（M）21例，重度（H）23例	鼾声、心电图和胸腹部偏移信号的采样熵	SVM	AUC：R vs. L：61%，R vs. M：68%，R vs.H：84%，L vs. M：63%，L vs. H：82%，M vs. H：65%
YE等^[22]	2024	正常9例，轻度OSA 11例，中度19例，重度55例	MFCC，LPC，SE， PR800，F₀，共振峰等	ReliefF-mRMR 特征筛选，SVM，LR，KNN，NB分类	SVM在AHI分别为5、15、30的阈值下，识别OSA准确度分别为100%、100%和95.94%
Luo等^[23]	2020	受试者132例，其中男109例，AHI为40.62±26.62，女23例，AHI为24.73±27.64	CNN提取的深度特征	逻辑回归	在AHI分别为c15、30的阈值下，分类准确度分别为80.17%和80.21%
Cho等^[24]	2022	受试者423例，AHI为32.6±24.4	LPC，MFCC，SE等	随机森林	在AHI分别为5、15、30的阈值下，识别OSA准确度分别为88.2%、82.3%和81.7%
Hou等^[25]	2021	正常10例，轻度OSA 23例，中度24例，重度36例	ECD	GMM	分类准确度达87.74%，其中正常、轻度、中度和重度的准确度分别为100%，81.48%，75.75%和93.75%
Qiu等^[26]	2024	正常26例，轻度OSA 45例，中度44例，重度82例	经PubMedBERT 转换的MFCC	XGBoost	分类准确度为58%，正常、轻度、中度和重度的准确度分别为50%、50%、66.7%和75%
Fang等^[27]	2025	正常40例，轻度OSA 40例，中度72例，重度313例	梅尔频谱图	AST	在AHI分别为5、15、30的阈值下，识别OSA准确度分别为92.6%、88.7%和83.0%
Le等^[28]	2023	正常282例，轻度OSA 316例，中度316例，重度401例	梅尔频谱图	DNN+线性回归模型	按AHI分别为5、15和30划分，敏感度为0.97、0.85、0.96，特异度为0.89、0.84、0.91
Romero等^[29]	2022	正常13例，轻度OSA 67例，中度38例，重度39例	梅尔滤波器组特征	DNN	AHI=15：敏感度79%，特异度80%； AHI=30：敏感度78%，特异度93%
Wang等^[30]	2022	训练组116例，AHI为27.6（9.9~61.6）；验证组19例，AHI为18.9（5.9~62.1）；测试组59例，AHI为24.4（6.6~54.3）	梅尔频谱图	OSAnet	在AHI=5、10，15、30的阈值下，识别OSA敏感度分别为93.6%、82.5%、88.5%和95.6%

作者	发表年份	样本分布	特征	模型	模型性能
Liu等^[31]	2022	受试者42例，AHI中位数为27.45	MFCC	KNN	模型准确度为85.55%，结合年龄、性别和BMI，准确度为87.98%，鄂后，舌后和多级阻塞分别为85.88%、89.22%和88.17%
Ding等^[32]	2022	阻塞部位：软腭484，口咽216，舌39，会厌89	梅尔频谱图	嵌入Conv6NP的原型网络	模型的非加权平均召回率为77.13%，其中软腭、口咽、舌、会厌分别为75.5%、80.0%、75.0%和78.0%
Sebastian等^[33]	2020	共58例患者，其中舌阻塞32例，侧壁11例，软腭10例，多级5例	时域特征：能量，熵，零交叉率；频域特征：共振峰频率，MFCC，光谱色度特征，基波和谐波频率	LDA	模型的总体准确度为62%，舌/非舌阻塞准确度为77%
Sun等^[34]	2020	阻塞部位：软腭484，口咽216，舌39，会厌39	TCC	特征筛选：主成分分析；分类：SVM	模型非加权平均召回率为87.5%，软腭、口咽、舌、会厌的准确度分别为93%，83%、74%和96%
Sun等^[35]	2021	阻塞部位：软腭484，口咽216，舌39，会厌89	MSE，MFCC	特征筛选：主成分分析；分类：SVM	模型准确度为89.09%，敏感度为86.36%，特异度为96.4%，其中软腭、口咽、舌、会厌的敏感度分别为98.04%、80.56%、72.73%和94.12%