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

Table 5

Comparison of WER results based on different datasets using the CNN-LSTM technique.

Features	CNN-LSTM technique
	DS1 (%)	DS2 (%)
MFCC 13 features	4.21	2.26
MFCC 25 features	4.04	1.26
MFCC 39 features	4.04	1.43
Filter bank 40 features	4.21	4.21
Filter bank 80 features	4.50	3.54
Filter bank 120 features	3.63	3.46

The bold values indicate the best comparison of WER results based on different datasets using the CNN-LSTM technique.