We are still working on the architecture for the decoder. The current decoder in Sphinx 3 combines a LexTree with an n−Gram language model. We feel this is too specific to include in sphinx4 and Sphincs, and are working on a flexible model that permits the Sphinx 3 model as well as other types of searches. At a minimum the other types of searches we are looking at include support for isolated word recognition, context free grammars, and n−gram language models.

The basic idea behind this model is to allow for more flexibility when it comes to the search strategy.

Last edited Oct 12, 2014 at 5:27 PM by DxN, version 3