Research On The Voice Conversion System

The research of voice conversion becomes a new field of study in speech signalprocessing after speaker recognition technology and speech synthesis technology currently,the research of voice conversion has a lot of theoretical and application value. However,the existing voice conversion is far from the hopes of people, there are still manyproblems to be solved for real application.Aiming at the current problems of voice conversion accuracy is not high enough andthe conversion algorithm is more complex, the characteristic parameters selection,characteristic parameters extraction algorithms and the conversion algorithms were mainlystudied in this paper.In the part of the selection of voice conversion parameters, by the research ofpersonality characteristics that affect human voice acoustic parameters such as pitch,formant, Mel frequency cepstrum experimental and theoretical analysis, identify the pitchand formant as the transformation parameters of voice conversion. These two parameterscan not only fully reflects the differences between the individuals, but also easy to extractand computing itself.In the part of the parameter extraction algorithms and transformation algorithms, bythe research of parameter extraction methods and transformation algorithms which arecommonly used, this paper proposes a method of spectral envelope conversion usingformant shift factor, the method is simple in calculation and the speech database requiredis relatively small. The voice conversion system simulation which successfully realizedthe voice conversion is carried out, which proved that the accuracy of the male to male canreach96%, the accuracy of the female to female can reach94%.