Construction of “Three Platforms and Six Sessions” Language Education Practice Model of Artificial Intelligence-Driven Industry-Teaching Integration

Abstract

The study derives from the educational practice model of “three platforms and six links”, synthesizes the multidimensional competence requirements, and uses two modules, speech recognition and language assessment, to drive the innovation of language education. The speech recognition module collects the speech of teacher-student interactions in the language teaching classroom, and realizes the adaptive recognition of lectures based on the Maximum A Posteriori Probability Reassessment (MAP) method, which combines the differential features of speakers. The language assessment module is grounded in bidirectional gated recurrent units and generative adversarial networks, and optimizes the decoding capability by improving the language audio feature encoder. In addition, the GAN with added LSTM units and fully connected layers can accomplish the discrimination of the authenticity of language evaluation more quickly. The language assessment method is able to adapt to different phoneme lengths in different language learning contexts, and when the phoneme length grows from 0 to 80, the assessment inference time only increases by less than 10ms, which has almost no effect. In the practical application, both teachers and students recognize the educational practice model of “three platforms and six links”, and the students' language proficiency scores have increased to different degrees.