Indonesian Lip-Reading Detection and Recognition Based on Lip Shape Using Face Mesh and Long-Term Recurrent Convolutional Network

Table 3

Performance results of IndoLR.

Model	Words			Rt (ms)	Phrase			Rt (ms)
	Val. acc. (%)	Test acc. (%)	Tt (s)		Val. acc. (%)	Test acc. (%)	Tt (s)
Conv-LSTM	94.7	90.42	7996.6	±88	95	90.62	9509.8	±105
LRCN-2Conv	95.83	92.92	600.3	±64	99.37	95.00	539.6	±68
LRCN-3Conv	97.92	95.42	727.3	±62	99.37	95.63	585.9	±66

Tt, training time in seconds; Rt, average recognition time for each video sample in milliseconds.