Monitoring of Sleep Breathing States Based on Audio Sensor Utilizing Mel-Scale Features in Home Healthcare

Table 2

Summary of previous studies on breathing states detection by acoustic signal.

Authors	Method	Features	Dataset	Snore detection results	Abnormal detection results
[23]	Voice activity detection algorithm	FFT	50 normal people breath 20 cycles and hold their breath to make the apnea	Not mentioned	Apnea detection accuracy more than 97%
[24]	CNN + RNN	CQT spectrogram	Part of full night recordings from 38 subjects	Accuracy: 95.3%	Not mentioned
[25]	CNN + LSTM	MFCC, LPCC and LPMFCC	Whole night recoding from 32 volunteers	Accuracy: 87%	Calculate AHI values
This study	Threshold values for individuals	Mel-scale-based features	Full nights recoding from 8 testers	Accuracy: 96.1%	Accuracy: 93.1%