Degree of confidence on big command phrase

[cornelyus]

Good morning.

I have been working with speech API for some weeks now regarding a new
project. We want command identification, rather than dictation.. the problem
is we have some large commands to identify.

For example.. the command "Julia wants to fly from London to New York" ..
this command i can divide into pieces like :

Subject Action City of Origin City of Destination

Sometimes my program identifies the whole phrase correctly, even if i say
something not on the grammar on the middle of it.. like the name of some city
not listed. My question is.. when the engine recognizes a phrase, it's degree
of confidence is given by a degree of the whole phrase, or each part of the
phrase has a high degree of confidence?

If it's the confidence on the whole phrase, wouldn't larger phrases have
more tendency for errors? Isn't there a way of wanting a high degree of
confidence on each part of the phrase?

Thanks for your time