Pitch Modification of Speech Residual Based on Parameterized Glottal Flow with Consideration of Approximation Error

Pitch modification plays an important part in the field of speech generation. For this task various approaches exist using time signals like PSOLA and related methods. Further approaches exist by using a parameterized model of the glottal flow or its derivative. In this contribution a hybrid approach is proposed based on a partly parameterized model of the glottal flow plus the consideration of the error of the model approximation; both of them are obtained from the residual of speech. By the consideration of the approximation error a high naturalness of the synthetic speech can be achieved; concurrently the use of the glottal flow model allows the alteration of pitch in a large range without appreciable decrease of speech quality