Improved Generation of Fundemental Frequency in HMM-Based Speech Synthesis Using Generation Process Model

The HMM-based Text-to-Speech System can produce high quality synthetic speech with flexible modeling of spectral and prosodic parameters. However the quality of synthetic speech degrades when feature vectors used in training are noisy. Among all noisy features, pitch tracking errors and corresponding flawed voiced/unvoiced (VU) decisions are the two key factors in voice quality problems. Pitch tracking errors occur more often in Mandarin vowels of Tone 3 and Tone 4. On the other hand, due to the dis-continuous F-0 values in voiced and unvoiced regions, it is then impossible to use standard HMMs for F-0 modeling. Currently a preferred method to solve this is to use a multi-space distribution HMM (MSD-HMM). In this approach, discrete distributions are used for modeling the VU decision and continuous Gaussian distributions are used for F-0 modeling within the voiced regions. Due to this assumption of undefined F-0 values in unvoiced regions and the special structure of MSDHMM, the generated F-0 values are limited in accuracy. In this paper, an F-0 generation process model is used to re-estimate F-0 values in the regions of pitch tracking errors, as well as in unvoiced regions. A prior knowledge of VU is imposed in each Mandarin phoneme and they are used for VU decision. Then the F-0 can be modeled within the standard HMM framework.