Table 5:

Transfer Learning Results (DNN: Hybrid DNN-HMM + offline i-vector AV: Acoustic Variability Modeling, PV: Pronunciation Variability Modeling) - 91 hours

Model	AV	PV	Configuration	WER
DNN Children	✗	✗	Baseline	25.53%
DNN Adult	✗	✗	Baseline	39.32%
DNN Children + Adult	✗	✗	-	20.35%
DNN TL	✗	✓	1 layer	26.97%
DNN TL	✓	✗	1 layer	24.26%
DNN TL	✓	✓	1 layer each	19.63%
DNN TL	✓	✓	dis-joint 1 layer each	20.01%
DNN TL	✓	✓	2 layers each	17.8%
DNN TL	✓	✓	dis-joint 2 layers each	18.74%
DNN TL	-	-	all layers	17.8%