HMM-based singing voice synthesis system using pitch-shifted pseudo training data

Over the last few years, a statistical parametric approach to singing voice synthesis based on hidden Markov models (HMMs) has been grown over. In this approach, spectrum, excitation, and duration of singing voices are simultaneously modeled by context-dependent HMMs, and waveforms are generated from HMMs themselves. However, pitches that rarely appear in training data cannot be properly generated because the system cannot model their fundamental frequency (F-0) contours. In this paper, we propose a technique for training HMMs using pitch-shifted pseudo data. Subjective listening test results show that the proposed technique improves the naturalness of the synthesized singing voices.