The protection and inheritance of Chinese language and writing intangible cultural heritage is facing the urgent demand of digital transformation, this paper proposes a speech recognition algorithm that integrates metalearning and transfer learning. The MetaTL algorithm captures the common patterns of cross-language speech features through meta-measure learning network, and combines with the dynamic gradient strategy to realize the fast adaptation of the target language. Relying on performance tests and functional tests, the effectiveness of the proposed algorithm is evaluated. IPA analysis is used to examine the practical application effect of the algorithm. The experiments show that the MetaTL algorithm converges after 111 rounds of training, reaching a correctness rate of 0.9 (90%), and the loss value stabilizes below 0.4 after 10 training sessions. The results of the survey show an overall mean importance value of 3.965 and a mean satisfaction value of 3.757. The interface dimension is located in the first quadrant “Advantageous Enhancement Zone”, the system functionality dimension is located in the second quadrant “Continuation Zone”, and the recommendation intention dimension is located in the third quadrant “Subsequent Opportunity Zone”. The dimension of system functions is located in the second quadrant “continue to maintain”, the dimension of willingness to recommend is located in the third quadrant “follow-up opportunities”, and the dimension of user experience is located in the fourth quadrant “urgent need to improve”. This study provides technical feasibility and practical paths for the digital preservation of linguistic non-heritage, and promotes the in-depth application of AI technology in the field of cultural heritage.
