This work extends and improves a recently introduced (Dec. 2007) dynamic Bayesian network (DBN) based audio-visual automatic speech recognition (AVASR) system. That system models the audio and visual components of speech as being composed of the same sub-word units when, in fact, this is not psycholinguistically true. We extend the system to model the audio and visual streams as being composed of separate, yet related, sub-word units. We also introduce a novel stream weighting structure incorporated into the model itself In recognition accuracy in a large vocabulary continuous speech recognition task (LVCSR). The "best" performing proposed system attains a WER of 66.71% whereas the "best" baseline system performs at a WER of 64.30%. The proposed system also improves accuracy to 45.95%from 39.40%.