SYNTHESIZING VISUAL SPEECH TRAJECTORY WITH MINIMUM GENERATION ERROR

In this paper, we propose a minimum generation error (MGE) training method to refine the audio-visual HMM to improve visual speech trajectory synthesis. Compared with the traditional maximum likelihood (ML) estimation, the proposed MGE training explicitly optimizes the quality of generated visual speech trajectory, where the audio-visual HMM modeling is jointly refined by using a heuristic method to find the optimal state alignment and a probabilistic descent algorithm to optimize the model parameters under the MGE criterion. In objective evaluation, compared with the ML-based method, the proposed MGE-based method achieves consistent improvement in the mean square error reduction, correlation increase, and recovery of global variance. It also improves the naturalness and audio-visual consistency perceptually in the subjective test.

PhotoReal_MGE_ICASSP2011_v3.pdf
PDF file

In  ICASSP 2011

Publisher  IEEE

Details

TypeInproceedings
> Publications > SYNTHESIZING VISUAL SPEECH TRAJECTORY WITH MINIMUM GENERATION ERROR