Search results for: lipreading

Authors: Peng Wang, Tigang Jiang

Abstract:

The computer based lip reading method of different languages cannot be universal. At present, for the research of Chinese lip reading, whether the work on data sets or recognition algorithms, is far from mature. In this paper, we study the Chinese lipreading method based on machine learning, and propose a Chinese Sentence-level lip-reading network (CNLipNet) model which consists of spatio-temporal convolutional neural network(CNN), recurrent neural network(RNN) and Connectionist Temporal Classification (CTC) loss function. This model can map variable-length sequence of video frames to Chinese Pinyin sequence and is trained end-to-end. More over, We create CNLRS, a Chinese Lipreading Dataset, which contains 5948 samples and can be shared through github. The evaluation of CNLipNet on this dataset yielded a 41% word correct rate and a 70.6% character correct rate. This evaluation result is far superior to the professional human lip readers, indicating that CNLipNet performs well in lipreading.

Keywords: lipreading, machine learning, spatio-temporal, convolutional neural network, recurrent neural network

Procedia PDF Downloads 92

Authors: Heng Yang, Tao Luo, Yakun Zhang, Kai Wang, Wei Qin, Liang Xie, Ye Yan, Erwei Yin

Abstract:

Recent years have seen that audio-visual recognition has shown great potential in a strong noise environment. The existing method of audio-visual recognition has explored methods with ResNet and feature fusion. However, on the one hand, ResNet always occupies a large amount of memory resources, restricting the application in engineering. On the other hand, the feature merging also brings some interferences in a high noise environment. In order to solve the problems, we proposed an effective framework with bidirectional distillation. At first, in consideration of the good performance in extracting of features, we chose the light model, Efficientnet as our extractor of spatial features. Secondly, self-distillation was applied to learn more information from raw data. Finally, we proposed a bidirectional distillation in decision-level fusion. In more detail, our experimental results are based on a multi-model dataset from 24 volunteers. Eventually, the lipreading accuracy of our framework was increased by 2.3% compared with existing systems, and our framework made progress in audio-visual fusion in a high noise environment compared with the system of audio recognition without visual.

Keywords: lipreading, audio-visual, Efficientnet, distillation

Procedia PDF Downloads 97