A neural network model for Chinese speech synthesis
Zengjun Xiang, Guangguo Bi · 2002
A multilayer back-propagation neural-network model which performs Chinese speech synthesis with high speed quality is described. The following are detailed: (1) the basic properties of the neural network and its difference from the traditional methods for speech processing; (2) selection of basic acoustic units which can be concatenated to generate utterances; (3) selection of the formant parameters; and (4) hybrid formant synthesizer and its neural network architecture. The test written in Chinese phonetic symbols (Chinese pingyin) is conveyed into the system and the speech with unlimited vocabulary is produced by the rules.>