Artificial neural-network based multilingual text-to-speech synthesis using a single set of phonetic symbols

摘要

The present invention relates to an artificial neural network-based multilingual speech text speech synthesis method and apparatus using a single speech symbol set, comprising: generating a spectrogram by receiving speech data of a first language; receiving the text of the first language and converting it into a single voice symbol; receiving text in a second language and converting it into a single voice symbol; generating a text-to-speech synthesis model by learning the spectrogram and the data changed into the single speech symbol with an artificial neural network, and using the learned text-to-speech synthesis model to convert the second uttering the language.