Cross-stream dependency modeling using continuous F0 model for HMM-based speech synthesis

Abstract

In our previous work, we have presented a cross-stream dependency modeling method for hidden Markov model (HMM) based parametric speech synthesis. In this method, multi-space probability distribution (MSD) was adopted for F0 modeling and the voicing decision error influenced the accuracy of generated spectral features severely. Therefore, a cross-stream dependency modeling method using continuous F0 HMM (CF-HMM) is proposed in this paper to circumvent voicing decision during the generation of spectral features. Besides, in order to prevent over-fitting problem in model training, regression class is introduced to tie the transform matrices in dependency models. Experiments on proposed methods show both improvement on the accuracy of the generated spectral features and effectiveness of introducing regression class in dependency model training.