HQSYN16

(je to trochu víc než 250 znaků)

• In the HMM-based speech synthesis, the number of states is usually equal for all models. We performed experiments with a variable number of states, optimal for particular models.
• An experimental comparison of several speech analysis-synthesis methods (STRAIGHT, GlottHM, Ahocoder and WORLD) was performed.
• We also performed a comparison of using HMM and DNN in the statistical parametric speech synthesis framework.