Minimum generation error training by using original spectrum as reference for log spectral distortion measure
Yi-Jian Wu, Keiichi Tokuda · IEEE International Conference on Acoustics Speech and Signal Processing · 2009
This paper improves a minimum generation error (MGE) based HMM training technique for HMM-based speech synthesis by directly using the original spectrum instead of line spectral pairs (LSPs) as reference spectrum for log spectral distortion (LSD) measure. Two types of original reference spectra for LSD calculation are investigated, including the spectrum extracted from speech waveform by STRAIGHT, and the short-time FFT spectrum calculated from speech waveforms. Since only the harmonics of the FFT spectrum are coincident with the underlying spectral envelope, the LSD between generated LSPs and original FFT spectrum is calculated by sampling at the harmonic frequencies, and a weighting function is designed to simulate the sampling strategy on LSPs. From the experimental results, the MGE-LSD training using the FFT spectrum as reference spectrum achieved the best performance.
Cited by 12 works. doi:10.1109/icassp.2009.4960508
Authors
References
- A Theory of Adaptive Pattern Classifiers
- Restructuring speech representations using a pitch-adaptive time–frequency smoothing and an instantaneous-frequency-based F0 extraction: Possible role of a repetitive structure in sounds
- A robust parser for spoken language understanding
- Line spectrum representation of linear predictor coefficients of speech signals
- Speech parameter generation from HMM using dynamic features
- Hidden Markov models based on multi-space probability distribution for pitch pattern modeling
- Speech synthesis using HMMs with dynamic features
- Speech synthesis from HMMs using dynamic features
- Minimum Generation Error Training for HMM-Based Speech Synthesis
- Speaker-Independent HMM-based Speech Synthesis System: HTS-2007 System for the Blizzard Challenge 2007