Phonetic conversion method, device and system

Abstract

Embodiments of the present disclosure relate to a word pronunciation conversion method, device and system. The method comprises: obtaining the maximum number of mapped phonemes corresponding to each letter included in each word in a set of test words, and a set of mapped phonemes corresponding to each letter; copying each letter to form a letter combination and generating position data corresponding to each letter in the letter combination; generating fusion features according to each letter combination and each position data; performing feature learning on the fusion features using at least two neural network systems to obtain a set of phonemes and a probability corresponding to each phoneme in the set of phonemes; and predicting a target set of phonemes corresponding to each word in the set of test words according to each word, the set of phonemes corresponding to each letter, the probability corresponding to each phoneme in the set of phonemes, and the set of mapped phonemes corresponding to each letter. Through the method, the accuracy of word pronunciation conversion is greatly improved.

Images