Output details
11 - Computer Science and Informatics
University of Edinburgh
Speech-driven lip motion generation with a trajectory HMM
<22> Originality: Investigation into the optimal model unit for lip synchronisation with speech for speech-driven animated agents based on statistical models. The first system using trajectory hidden Markov models to automatically map audio speech features to lip motions without using heuristic smoothing filters.
Significance: The objective and subjective comparisons done in the study revealed the quality of synthesised lip motions depends on the model unit for lip motions, i.e. visemes, and suggested optimal selection of unit is possible rather than defining the unit heuristically.
Rigour: Developed 3D motion capture data for speech-driven lip synchronisation.