Voice conversion algorithm based on Gaussian mixture model applied to STRAIGHT

摘要

As a typical voice conversion algolithm, the codebook mapping method has been studied by Abe et al. The main shortcoming of this method is the fact that the acoustic space of a speaker is limited to a discrete representation. To represent it continuously, the algolithm based on Gaussian mixture model has also been proposed by Stylianou et al. In this paper, we apply this algorithm to STRAIGHT proposed by Kawahara et al, which is recognized as a high quality vocoder. For evaluating this conversion algolithm, evaluation experiments were performed by comparing with the algolithm based on the codebook mapping method. As a result, a performance of the algorithm based on Gaussian mixture model was better than the algorithm based on the codebook mapping method. Also, effects by the amounts of training data for the conversion algorithms were investigated, as well as the number of the Gaussian mixtures.