| RESPITE:Events : Meeting, Sep 2000:Presentations: Andrew Morris |
In HMM/ANN hybrid ASR there are some situations where MLP experts cannot be used because they do not provide a model of the data likelihood. GMM based experts can provide both posterior probabilities and data likelihoods. Several applications of the HMM/GME hybrid are discussed: recognition with missing data (position of MD known/unknown), and adaptive multi-stream expert weighting (ML and clean likelihood based).