Table 3.

Segmentation and Feature Extraction Strategies.

Author (Year)	Event Stamp	Segmentation Methods	Feature Extraction Strategy/Source
Afkari [30]	Ev	Manual segmentation	Time domain raw signal
Amft and Troster [31]	Ep	Frame at 250 ms	Feature Similarity Instance
Bi et al. [32]	Ev	HMM-based on Mel frequency cepstral coefficients	Predetermined time-domain, frequency-domain, and non-linear features
Fontana et al. [33]	Ep	Frame at 1.0 s & 1.5 s	Time domain raw signal
Fukuike et al. [34]	Ev	Identifying the semblable wave period by moving average. A period longer than 0.35 s was regarded as a swallowing event	Time domain raw signal
Kurihara et al. [35]	Ev	Manual prepared template for pattern matching
Lee et al. [36]	Ep	Frame at 200 ms with 50% overlap	Signal variance
Makeyev et al. [37]	Ev & Ep	1.5 s epoch	Mel-scale Fourier spectrum with PCA
Sazonov et al. [38]	Ev & Ep	Grid search on epoch duration and step size	Frequency domain raw signal
Sejdic et al. [39]	Ev	Minimum Description Length-based Sequential Segmentation	Time domain raw signal
Skowronski et al. [40]	Ep	Manually segmentation at 6 s	Human factor cepstral coefficients and spectral flatness measure

Ep: Epoch-based; Ev: Event-based; HMM: Hidden Markov Model; PCA: Principal Component Analysis.