Figure 6
From: SynFace—Speech-Driven Facial Animation for Virtual Speech-Reading Support
Structure of the language models mappers. (a) Phonetic mapper: only identical phonemes are matched between the two languages, (b) best match mapper: the phoneme are blindly matched in a way that recognition results are maximised, (c) linear regression mapper: posterior probabilities for the target language are estimated from the outputs of the phonetic recogniser.