Table A2.

Training results for different deep learning architectures and varying window sizes between one and two seconds. All models were trained and optimized as described in Section 3.

	Windowsize in Samples
Architecture	200	300	400
CNN	0.884	0.893	0.887
LSTM	0.867	0.860	0.830
convLSTM	0.852	0.853	0.857