Voice conversion method for e.g. text-to-speech system, involves transferring set of prediction-live prediction code-coefficients for voice conversion with manipulated stimulation signals of speech synthesis filter during voice synthesis

摘要

The method involves providing a speech synthesis filter (7) for generation of voice stimulation signals and line spectrum frequencies (LSF)-parameter. A modified set of LSF-parameters are computed from a set of decoded LSF-parameters by transformation relationship. A manipulated set of LSF-parameters are converted to a set of prediction-live prediction code (LPC)-coefficients. The set of prediction-LPC-coefficients is transferred for voice conversion with manipulated stimulation signals of the filter in a processor-controlled manipulation unit (8) during voice synthesis. An independent claim is also included for a device for voice conversion by speech decoding and speech synthesis.