Table 4.

VAD frame-level cross-validation performance vs. model type–RLDD and RODeCAR.

Metric	Model
	Original CNN 127-pt. DFT, 8 kHz s.r. No Retraining (Baseline)		Original CNN 127-pt. DFT, 8 kHz s.r. Retrained (10%)		Original CNN 255-pt. DFT, 16 kHz s.r. Retrained (10%)		CNN-MLP 255-pt. DFT, 16 kHz s.r. Retrained (10%)
	RLDD	RODeCAR	RLDD	RODeCAR	RLDD	RODeCAR	RLDD	RODeCAR
Accuracy [%]	83.22	80.55	86.18	85.92	91.53	86.37	94.92	89.92
Precision [%]	89.55	87.34	–	–	–	–	96.79	90.89
FAR [%]	35.73	18.25	–	–	–	–	11.05	14.69
FRR [%]	11.31	20.21	–	–	–	–	3.35	7.15