A hybrid neural system for driving a parallel synthesiser so as to synthesise high quality accented speech

Investigations have been carried out into the use of an optimized computing structure to synthesize high quality accented speech. The computing structure forms a seven formant parallel speech synthesizer where each element of the structure comprises a neural network adjusting the outputs of a rule base. Results are presented as to the effectiveness of the hybrid rule based/rule following element, in comparison to a standard neural network, when modeling a single FET circuit and for the performance of both a cascade and global hybrid system, in comparison to a series configuration. When the hybrid structure is applied to the synthesis of accented speech the rule based part of the system models a seven formant parallel synthesizer and the neural network part of the structure alters the outputs from the rule bases. The network is then trained with accented speech