Research on the construction of acoustic model for foreign language speech recognition based on deep neural network

Aiming at the problem of insufficient accuracy of acoustic modeling in foreign language speech recognition, this paper constructs a BiLSTM acoustic model based on deep neural network, combining FBank feature extraction and multi-strategy hyperparameter optimization to improve the model’s ability of temporal modeling of foreign language speech features. The experiments are carried out on TIMIT and Common Voice corpus, and the results show that the constructed model is better than the traditional model in terms of recognition accuracy and stability, which verifies the effectiveness and practical value of the method.