In this paper, we propose a HMM-based text-to-speech (TTS) using sub-band basis spectrum model (SBM). SBM can represent vocal tract spectra and phase characteristics by liner combination of sub-band basis vectors. Some reports suggest that analysis-synthesized speech based on SBM is close to the natural speech and SBM can perform effectively in the text-to-speech. Therefore, SBM framework is expected to improve speech quality to have good effects on the HMM-based TTS. Subjective experimental results show that the proposed method improves speech quality in some conditions.

Index Terms: speech synthesis, hidden Markov model, sub-band basis spectrum model, phase feature