| In the past ten years,thanks to the sharp increase of data flow,the rapid improvement of computing capability and the emergence of deep learning,the wave of artificial intelligence has gradually swept the world.From image recognition to speech recognition,from robots to expert systems,the application of artificial intelligence technology is constantly being innovated and being integrated into our lives.As a hot spot of current Internet applications,the field of online education also needs the assistance of artificial intelligence technology.In recent years,artificial intelligence technology has many applications in the fields of text recognition and voice recognition.In the Optical Character Recognition(OCR)task,the traditional feature extraction method can effectively recognize document-like pictures,but it does not perform well in Scene Text Recognition(STR);at the same time,handwritten mathematical formula recognition is still a major difficulty in research.Therefore,how to apply deep learning to text recognition tasks is still a research hotspot.In addition,the development of neural networks has brought new ideas to voiceprint recognition and speech emotion recognition tasks.However,it is still the focus of research to make better use of it.In order to address the above issues,the thesis aims to use artificial intelligence technology to assist online education.It is specifically carried out from three aspects:First,for the text recognition task in the photo search scenario,deep learning-based CTPN and CRNN algorithms are investigated.The superiority of the deep learning algorithm in the STR task is verified through comparative experiments with traditional OCR.Subsequently,VGG was designed for the defect that CTPN+CRNN can't recognize oblique text.Secondly,for the task of handwritten mathematical formula recognition,we investigated and reproduced the WAP algorithm based on deep learning,and then added a mathematical grammar correction module to improve the accuracy of recognition.Finally,for the teaching attitude evaluation task,the Deep Speaker(voiceprint recognition algorithm)and VoiceFilter(voice separation algorithm)were investigated and implemented respectively,and then a deep learning-based speech emotion recognition model was designed and implemented to complete the quantitative evaluation of the teacher's teaching attitude. |
PDF Full Text Request