The present invention provides a method and an apparatus for generating a grapheme-to-phoneme conversion model on the basis of an artificial intelligence. The method comprises the following steps: obtaining a retaining node for training each word by conducting a random screening for a hidden hierarchy node in a neural network as a proportion of a preset node, in every training process for performing a grapheme-to-phoneme conversion for each word of training data by applying the neural network; training each word by applying a sub-neural network corresponding to the retaining node, and updating relevant weight of each retaining node in the sub-neural network; and producing a grapheme-to-phoneme conversion model by finally conducting an average processing for the relevant weigh of each retaining node in the overall sub-neural network. Through the method and apparatus for generating a grapheme-to-phoneme conversion model on the basis of an artificial intelligence, a learning ability and a generalizing ability of the grapheme-to-phoneme conversion model are improved, and an over-fitting phenomenon can be avoided.
展开▼