CCA provides two modes: Free-speaking mode and Command-picking mode.
User can speak any thing. What she/he said will be converted into text and be sent to the network in real time. The disadvantage of this mode is the recognize precise is not high.
A command list($cca_path/data/commandList.txt) which contains all available commands is required. Anything you spoke will be converted into the most similar command in the command list.
For example, if you edit commandList.txt as:
one two three
nine one one
Then, any audio will be recognized as "one two three" or "nine one one".
Command-picking mode seems not so "free" as Free-speaking mode, but it can give a high recognize precise.