Research Article

A Chinese Lip-Reading System Based on Convolutional Block Attention Module

Table 3

Testing accuracy of each model.

Model/characterResNet101 + LSTM + Attention (%)ResNet101 + GRU + Attention (%)ResNet152 + LSTM + Attention (%)ResNet152 + GRU + Attention (%)ResNet50 + CBAM + LSTM + Attention (%)ResNet50 + CBAM + GRU + Attention (%)

Ling (Zero)98.0016.0085.0061.0099.0097.00
Yi (One)93.0099.0072.0099.0056.0098.00
Er (Two)99.0099.0099.0099.0099.0099.00
San (Three)99.0096.0098.0099.0086.0099.00
Si (Four)70.0099.0092.0035.0073.0070.00
Wu (Five)99.0099.0098.0099.0099.0097.00
Liu (Six)99.0099.0099.0099.0099.0098.00
Qi (Seven)95.0070.0085.0076.0097.0098.00
Ba (Eight)98.0099.0098.0099.0099.0099.00
Jiu (Nine)63.0074.0016.0099.0098.0097.00
Chi-Fan (Eat)97.0099.0060.0099.0078.0099.00
Dui-Bu-Qi (Sorry)95.0099.0099.0099.0063.0099.00
Ni-Hao (Hello)99.0099.0099.0099.0099.0099.00
Pao-Bu (Run)99.0099.0099.0099.0092.0096.00
Shui-Jiao (Sleep)99.0099.0073.0056.0041.0099.00
Wan-Shua (Play)99.0099.0099.0099.0099.0099.00
Xue-Xiao (School)99.0099.0099.0098.0093.0099.00
Zai-Jian (Good-Bye)45.0033.0092.0099.0086.0099.00
Zhong-Guo (China)99.0099.0099.0099.0098.0099.00
Zou-Lu (Walk)93.0099.0088.0097.0097.0099.00