Convolutional Neural Networks to Facilitate the Continuous Recognition of Arabic Speech with Independent Speakers

Table 3

WER and CER results of CNN-LSTM using filter.

No. of filter bank features	WER		CER
	DS1 (%)	DS2 (%)	DS1 (%)	DS2 (%)
40 features	4.21	4.21	4.42	2.98
80 features	4.50	3.54	4.12	1.27
120 features	3.63	3.46	4.05	1.82

The bold values indicate the best results of CNN-LSTM using filter bank from WER and CER.