In the standard approach the recognized word sequence is represented by the best scoring path in the word lattice. Within the presented approach, the recognized word sequence is determined by the best scoring path in a log-linear combination of the word lattices for every system.
Since the influence of the applied systems is crucial for the success of this method. Discriminative methods are utilized to estimate weights for each system. These weights are used to handle the effect of the individual systems on the final result.