International Workshop on Models and Analysis of Vocal Emissions for Biomedical Applications (MAVEBA 1999)
Abstract. The task of joint speech/video processing is considered. The approach based on two sets of autoregressive hidden Markov models (audio and video models) and neural network is proposed in order to improve the speech recognition performance. The data from each word is processed in two separate channels, and we have two sets of M aposteriory probabilities as the outputs. To combine these results in order to improve the processing accuracy we introduce the direct links neural network. Such technique can be especially useful for persons with limited physical possibilities.
Bibliographic reference. Bovbel, Evgeny I. / Kukharchik, P. D. / Kheidorov, Igor E. (1999): "The joint speech/video signal processing for persons with limited physical possibilities", In MAVEBA-1999, 144-145.