Overview of Typical Development Schemes

Decisions about the architecture of the recognizer can be made at many different stages of the development. On this page we will list the so far preferred default development scheme as well as variations and show when they are useful and what their consequences are. Every list entry will have a link to a page, where the single steps of the development scheme are described only very briefly, but each step has a link into all three threads with more detailed discussions and scripts and instructions for how to do it yourself.

Some of the items in the list below are not entire recognizer developments, but must be understood as parts of an entire development process, or as alternatives to other parts. But this will be clear after reading what the scheme is about.

The Standard Scheme
Provided you are not missing any important and helpful things like labels and a dictionary etc., this standard scheme will eventually produce a context dependent continuous (or three-quarter continuous) Gaussian mixture single LDA stream HMM recognizer.
Codebook Growing Scheme
We apologize that this development scheme is not yet documented.
A Semi-Continuous Approach
We apologize that this development scheme is not yet documented.
Something Else
We apologize that this development scheme is not yet documented.