Application Of Tibetan Speech Recognition Based On Active Learning In Online Education

With the development of science and information technology and internet technology,online education plays an important role in China's national education by its advantages of striding over time and space,sharing of high-quality educational resources.However,due to geographical and historical factors,insufficient educational resources Tibetan areas of China and lacks of excellent educational resources in the coastal areas of eastern central China.The sharing of high-quality education resources contributed by online education and the characteristics with spanning time and space can,to a certain extent,address the shortage of educational resources in Tibetan ethnic areas,and narrow the educational gap between the developed coastal areas and central and eastern areas in China.This results in the educational equality and accelerates the process of educational modernization in Tibetan areas.The speech recognition applied for teaching videos in the online education platform is a necessary module of video structured processing in the network.However,in China,the online education platform represented by XueTangX,Net Ease Open Class,China MOOC Network,Khan Academy,etc.is mainly spoken in Mandarin and English,and Tibetan-based speech recognition is relatively scarce.As Tibetan language is one of minor languages in China,most existing Tibetan speech recognition models use supervised learning methods to establish recognition models.In order to establish a high-accuracy speech recognition model,supervised learning requires a large amount of annotated speech corpora.But since the lack of linguistic annotation,extremely time-consuming and laborious task.In this thesis,we adopt a method based on active learning to select a small set of valuable samples from a large number of unlabeled speech data for users to label,so as to use a small amount of high-quality training samples to construct a recognition model as accurates as the model based on a large amount of training data.To reduce the amount of manually annotated data,improve work efficiency,solve the tedious and lengthy task of labeling,and accelerate the process of online education in Tibetan areas.This thesis explains the basic principles of active learning and Tibetan speech recognition,and the necessity and feasibility of applying them to online education videos in Tibetan areas.This thesis introduces the basic principles of active learning and Tibetan speech recognition and discusses on the necessity and feasibility of applying them to education video online in Tibetan ethnic areas.According to the phonetics knowledge of Tibetan language,we carry out the near optimal active learning based Tibetan speech recognition research,and then use QT Creator to build Tibetan speech recognition system based on active learning and speech-to-text automatic system for online education video.It realizes the real-time Tibetan speech recognition,meanwhile the recognition result with Tibetan characters is shown below the video.The formation of the teaching video with subtitles,deepens the learners deep understanding to the teaching video content,improve the learners to understanding the content of teaching video,improves the learners,learning efficiency,promotes the quality education in Tibetan areas of teaching resources,and promote the process of education informatization.